People's Republic of Bangladesh

Coastal Climate Resilience Infrastructure Project

(CCRIP)

Authors:

Aslihan Arslan

Daniel Higgins

Abu Hayat Md. Saiful Islam

The opinions expressed in this publication are those of the authors and do not necessarily represent those of the

International Fund for Agricultural Development (IFAD). The designations employed and the presentation of

material in this publication do not imply the expression of any opinion whatsoever on the part of IFAD concerning

the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its

frontiers or boundaries. The designations “developed” and “developing” countries are intended for statistical

convenience and do not necessarily express a judgement about the stage reached in the development process by a

particular country or area.

This publication or any part thereof may be reproduced without prior permission from IFAD, provided that the

publication or extract therefrom reproduced is attributed to IFAD and the title of this publication is stated in any

publication and that a copy thereof is sent to IFAD.

Arslan, A., Higgins, D. and Islam, A.H.M.S. 2019. Impact assessment report: Coastal Climate Resilience

Infrastructure Project (CCRIP), People's Republic of Bangladesh. IFAD: Rome, Italy.

Cover image: ©IFAD/GMB Akash

© IFAD 2019

All rights reserved.

1

Acknowledgements

The authors would like to thank all of those at the Local Government Engineering Department and

IFAD who assisted with the design and implementation of this impact assessment and provided

inputs for this report. This especially includes the CCRIP Project Director, Luthfur Rahman, and the

MEK Specialist, Shahjahan Miah, without whom this work would not have been possible, as well as

the GIS Specialist Neamul Ahsan Khan. From IFAD, we especially thank Peter Brückmann for his

extensive assistance during the data collection stage, David Hughes and Michelle Latham for their

extensive GIS support, and Philipp Baumgartner, Christa Ketting and Sherina Tabassum for their

support throughout.

Finally, we acknowledge the company hired to collect the data, BETS Consulting Services, along

with the sampled households themselves for their time and patience.

2

Table of Contents

Acknowledgements .......................................................................................................... 1

Executive summary .......................................................................................................... 3

Introduction ...................................................................................................................... 5

1. Project details, theory of change and main research questions .................................... 7

a. Project implementation ............................................................................................................... 7

b. CCRIP Theory of Change .......................................................................................................... 8

c. Research questions ..................................................................................................................... 12

2. Impact assessment design: Data and methodology .................................................... 13

a. Overall approach ........................................................................................................................ 13

b. Data ............................................................................................................................................... 14

c. Questionnaire and impact indicators ...................................................................................... 20

d. Impact estimation ....................................................................................................................... 25

3. Profile of the household questionnaire sample ........................................................... 29

a. Household characteristics by division ................................................................................... 29

b. Household characteristics by catchment area ...................................................................... 34

4. Results ........................................................................................................................ 36

a. Overall impacts of CCRIP ....................................................................................................... 36

b. Impact heterogeneity ................................................................................................................. 46

5. Conclusion .................................................................................................................. 51

References ...................................................................................................................... 53

Appendix I: Mean values for impact indicators ............................................................. 57

Appendix II: Distribution of Propensity Scores before and after trimming ................... 61

Appendix III: Results from the secondary IPWRA model ............................................. 62

3

Executive summary

The Coastal Climate Resilient Infrastructure Project (CCRIP) is a $150 million rural infrastructure

project which was implemented in 12 districts of Bangladesh since 2013, and is due to be completed

by the end of 2019. The project is funded by IFAD, the ADB, KfW of Germany, and the Government

of Bangladesh. The project aims to improve the connectivity of farms and households in the face of

climatic shocks, focusing on one of the most shock-prone areas of one of the most shock-prone

countries in the world. The main component of the project is the construction of improved markets

and market connecting roads, that are designed to remain useable during the monsoon season. This is

expected to improve sales of on-farm produce, along with access to inputs as well as opportunities for

off-farm income generation, leading to increased productivity and income. The project also aims to

improve women's empowerment by employing Labour Contracting Societies (LCS), consisting mainly

of destitute women, to carry out some of the construction work.

This impact assessment focuses on the activities funded by IFAD, which includes the strengthening of

markets and roads at the community and village levels. Using data from an in-depth household

questionnaire covering 3,000 treatment and control households, combined with extensive qualitative

interviews, we analyse the project's impact on a range of impact indicators relating to income; crop,

fish and livestock production and sales; assets, food security and education; financial inclusion; and

women's empowerment. We assess impact on the whole sample, as well as for a range of sub-groups,

including by geographic location, location within the market catchment area, and by livelihood

activity, integrating findings from the qualitative data to help to explain the mechanisms that shaped

the project's impact.

Regarding on-farm activities, we find that, despite a lack of impact on productivity, income from

selling crops and fish increased significantly (by 104 and 50 per cent, respectively). However, we do

not find a similar increase in income from the sale of livestock and livestock products. The lack of

impact on productivity was seemingly caused by persisting issues with accessing high-quality inputs

during the monsoon season, as well as households having limited capital to purchase these inputs.

Despite this, the project increased the amount of produce that was sold, the amount that was sold at a

market rather than from home or the farm gate, and increased the likelihood of growing cash crops,

leading to the large increase in on-farm income.

As well as improving on-farm income, the project also increased income from wage labour, which

together produced a positive impact on total income of 11 per cent, along with a four per cent

reduction in poverty. This increased prosperity was also reflected in reduced food insecurity and

increased ownership of households assets.

The project was intended to improve women's income generation and standing in the community,

mainly through its work with LCS, but we do not find an impact on women's contribution to

household income, or on their involvement in household decision making. When we analyse data

separately for Muslim and non-Muslim housheolds, we find that the project did improve these

indicators for non-Muslim households, suggesting that there were specific barriers faced by women

in Muslim households that the project was unable to overcome.

4

The additional sub-group analyses provided a number of additional insights. First, we find variation

in impact according to project district, which is caused by different impacts on income from crop

sales, livestock rearing, and wage employment. For the districts in Dhaka, for example, there was a

large positive impact on income of 21 per cent, which was driven by a large impact on livestock

income, which wasn't achieved in other areas. We also find that within the catchment areas of each

market, the larger impacts on income and poverty were achieved for the poorer, more remote

households. Finally, we find that the overall impact of the project was driven by improvements for

farming households (i.e. those with crop, fish or livestock production), while non-farming

households did not benefit significantly from the project.

Based on the findings of this impact assessment, we draw a number of important lessons for future

projects and policies that address the connectivity problem. First, future projects should pay special

attention to ensuring households have access to high-quality agricultural inputs, as improved access

to output markets does not necessarily improve access to inputs especially for credit constrained

households. This would help to increase productivity and thus stimulate larger impacts on income. In

addition to improving input access, agricultural productivity and income could also be improved by

providing complementary training and agricultural technology support.

Second, future projects should consider different components of beneficiaries' livelihoods and

provide activities to stimulate the main sources of income, which may vary for different areas. In

some cases for CCRIP, there was a lack of impact on the main income sources for some households

(mainly wage labour and household enterprises), leading to a lack of impact on total income. Impact

on income could thus be enhanced by offering complementary support for the livelihood activities

that are the most important in each local context. In the case of wage labour, this could involve a

redesign of the LCS activities to ensure that the valuable employment and training provided does not

remain short-term in nature and includes support to establish linkages with local labor market for

sustained impacts. In terms of household enterprises, these activities could be improved by

facilitating easier entry into local markets for small shops and traders, as well as providing credit and

training for setting up and managing these businesses.

Finally, future projects should provide more extensive support to improve the income generating

opportunities and overall empowerment of women, especially in countries such as Bangladesh where

women face ingrained barriers to their mobililty and autonomy. Based on the success of initiatives

such as BRAC's Empowerment and Livelihood for Adolescents program in Bangladesh (as well as

Afghanistan, Haiti, Sierra Leone, South Sudan, Tanzania and Uganda) future projects could provide

multi-faceted support to improve the hard and soft skills of women, provided within a safe space

environment, and involve the wider society to ensure sustainability of impacts.

5

Introduction

The Coastal Climate Resilient Infrastructure Project (CCRIP) is an inter-agency infrastructure

development intervention located in three divisions of southwest Bangladesh. The project

implementation started in 2013 and is due to be completed in 2019. It is funded with a combined

US$150 million from IFAD, the Asian Development Bank (ADB), KfW of Germany, and the

Government of Bangladesh, and is being implemented by a team from the Local Government

Engineering Department (LGED). In 2018, a combined data collection exercise was conducted by the

project team and a team from the Research and Impact Assessment (RIA) Division of IFAD to be used

for both the project's Mid-Term Report and an impact assessment study. This report details the design

and the findings of the impact assessment study.

CCRIP aims to improve the connectivity of farms and households in the face of climatic shocks,

focusing on one of the most shock-prone areas of one of the most shock-prone countries in the world

(Saha, 2014; Kreft, 2017). The project has three broad components: (i) Improved roads; (ii) Improved

market access; and (iii) Enhanced climate change adaptation capacity. One common theme across the

IFAD-funded components is their involvement of Labour Contracting Societies (LCS), which are

groups of mainly destitute women. These groups are contracted to carry out construction work, and

some of them are also provided with Women's Market Sections installed in community markets.

Upon completion, the project aims to reach 600,000 households from 32 Upazilas across 12 coastal

districts in the country. CCRIP was formed from the merging of IFAD's Sustainable Infrastructure for

Livelihoods Enhancement (SMILE) project with the ADB and KfW's Climate Resilient Infrastructure

Improvement in Coastal Zone Project (CRIICZP). This impact assessment focuses only on the impact

of the activities funded by IFAD, which consisted of the construction of climate resilient community,

union and village roads and markets, and the use and support of LCS.

In this impact assessment we test the impact of the project on a set of relevant impact indicators using

rigourous impact assessment methods, involving both quantitative and qualitative data and a carefully

constructed comparison group. This assessment has a number of benefits. Firstly, the insights from

this analysis help further understand how this type of project is expected to impact beneficiaries and

the contextual factors and barriers that can shape impact. Such insights can help to improve future

projects that seek to improve rural livelihoods in the face of climatic shocks, which are increasing and

are threatening rural poverty reduction worldwide (Kirtman et al., 2013; World Bank, 2017).

The second benefit of this analysis is that it contributes to IFAD's mandate to increase the

accountability of development spending. Along with 17 other projects, CCRIP was selected as part of

the IFAD10 Impact Assessment Initiative. Following on from the IFAD9 Impact Assessment

Initiative, the RIA division will use this set of impact assessments to extrapolate and estimate the

impact of IFAD's overall portfolio for its 10th replenishment period (2016-2018) (Garbero, 2016). This

initiative provides one of the most robust investigations into the impact of a development institution's

portfolio, and thus generates reliable insights into the results of IFAD's work and investments.

The final benefit of this work comes from its collaborative nature. By combining the work of the

project team and RIA, this assessment grants the opportunity to connect IFAD's field operations with

its impact assessment programme, creating a multi-stakeholder approach that facilitates insights that

6

are of the highest relevance and usefulness. In collecting detailed quantitative and qualitative data that

can be used to fulfil multiple reporting requirements, the work also provides an example of conducting

rigorous research on project performance in a cost-efficient manner.

The remainder of this report is structured as follows: Section 1 provides an overview of the context in

which the project was implemented, outlines the Theory of Change of the project, from which we

select our impact indicators, and presents the main research questions; Section 2 provides details of

the study methodology, the data, and the impact indicators analysed; Section 3 provides a profile of

the households included in the sample; Section 4 contains the results and discussion of the project's

impact; and Section 5 concludes with policy and programmatic implications.

7

1. Project details, theory of change and main research

questions

a. Project implementation

CCRIP effectively functions as three separate but conceptually linked sub-projects. IFAD's

component focuses on union and village roads and bridges, and on community and village markets;

while the ADB component focuses on larger scale Upazila roads, and large markets and growth

centers; and KfW focuses on the provision of cyclone shelters and other climate resilience support.

The IFAD interventions are being implemented in 32 Upazilas of 12 districts in southwest coastal

Bandladesh. Table 1 presents the CCRIP districts and the spread of project Upazilas across these

districts. These were identified from a set of 77 Upazilas that were assessed for inclusion using a

scoring system, which resulted mainly in the prioritisation of coastal, flood-prone, low-lying, and

infrastructure-poor chars. The scoring system was based on the following criteria:

Proportion of population below the

poverty line

Low wages for farm labour

Vulnerability to tidal surges,

storms, floods and river erosion

Remoteness

Poor communication (per cent of

paved road to total road)

Road density by population

Per cent of undeveloped markets

Table 1: Distribution of Upazilas across project districts

District Nr. Upazilas District Nr. Upazilas

Bagerhat 2 Khulna 3

Barisal 3 Madaripur 2

Bhola 3 Patuakhali 5

Borguna 4 Pirojpur 2

Gopalgonj 2 Satkhira 3

Jhalkati 1 Shariatpur 2

Within selected Upazilas, IFAD roads and markets are placed in areas that maximise the reach of

their benefits to poor people. This involves identifying the least developed unions and villages

within each Upazila, especially rural markets from char, low-lying, disaster-prone, and infrastructure

poor villages. For the LCS groups, households apply to be members, and are then selected based on

their poverty levels and experience in either construction or running a market stall.

8

For markets to be eligible for CCRIP support, they must meet the following criteria:

Strategically located and serve as an assembly market to benefit a large number of

villages and connect other larger market and growth centers;

Location not vulnerable to river erosion in the short and medium term;

Has potential for development in terms of availability of space and placing suitable

layouts;

Support from market stakeholders;

Agreement to share lease income with Market Management Committee.

Within the markets that meet this criteria, the final beneficiary markets are selected based on their

potential for poor women to participate in the construction of the market and as buyers and sellers;

the willingness of stakeholders to share part of the development cost to be used for the further

expansion of the works; and willingness of stakeholders to reserve sections for temporary sellers,

especially women and smallholders.

b. CCRIP Theory of Change

CCRIP is trying to solve a fundamental problem of rural development in southwest Bangladesh: the

low connectivity of smallholders' farms and households to markets, roads and urban centers (Rahman

and Rahman, 2015). The project has a particular focus on households living in char areas, which are

areas of land created by river sediment formed into sandbars along river channels, and are especially

remote and vulnerable to extreme weather shocks (Islam et al., 2014).

Low connectivity of households and farms hinders access to education, healthcare, financial and

support services, as well as employment oportunities. It also constrains access to input and output

markets, technology and productive facilities, and market information and extension services. This

lack of access has significant livelihood implications. At the household-level, limited access to these

services is widely regarded to negatively affect short and long-term livelihood quality and wellbeing,

including household food security and nutrition (Alkire and Santos, 2010,; Sibhatu et al., 2015;

Koppmair et al., 2017; Islam et al., 2018). At the farm-level, restricted access to input, technology,

extension and financial services can hinder the volume, quality and diversity of production, and

integration into value chains (Fan et al., 2012; Rehima et al., 2013; Bokelmann and Adamseged,

2016). Combined with poor access to vibrant markets and market information, plus high transport

costs, this can have a negative effect on the prices and profits that farmers receive for their goods

(FAO, 2003).

Both regular and unexpected climatic stresses exacerbate the connectivity issues in already remote

areas of southwest Bangladesh, especially in the char areas (Huq et al., 2015). During the annual rainy

season, many connecting roads become submerged and unusable, severely restricting transport. In

terms of unexpected shocks, the country experiences a tropical cyclone every three years, and a severe

flood every four-to-five years, with the southwest coastal region often bearing the brunt of the damage

(Nishat et al., 2013; Saha, 2014). For instance, two of the most recent major disasters in the country

damaged mainly the southwest region: Cyclone Sidr in 2007 caused 3,400 deaths and damaged

8,000km of roads; and Cyclone Aila in 2009 caused 180 deaths and damaged 7,000km of roads

(Relief Web, 2008; Relief Web, 2009).

9

CCRIP addresses the connectivity issue in the region by building and upgrading climate resilient roads

and markets. Roads are built to connect districts, villages and unions to each other and to markets.

These roads are made from materials that can withstand frequent submersion by salty or brackish

water. Roads are also raised and have higher and wider shoulders, with culverts and water gates

installed to manage flood water. Where suitable, vetiver grass is also used to line road slopes to

prevent erosion.

Households' market access is hindered by both a lack of transport infrastructure and a lack of physical

markets themselves. Cyclone Sidr alone is estimated to have caused damage and losses to the

country's agriculture sector of US$437 million, partly through damages to physical markets (Relief

Web, 2008). In order to complement the road work, CCRIP also establishes new markets and

upgrades existing ones. These markets range from "special" markets with over 200 permanent shops

serving over ten villages implemented by the ADB, to medium markets with around 100 permanent

shops serving up to ten villages, and smaller village markets with 10-50 shops serving up to four

villages implemented by IFAD. In terms of upgrades, CCRIP adds multi-purpose sheds, fish sheds,

boat landing platforms, open paved/raised areas, women's sections, toilet blocks, internal roads, and

improved drainage, depending on need.

The project also recognises that improving the management of markets in Bangladesh is key to market

sustainability (Ahmed, 2010). Market Management Committees (MMCs) are groups made up of

market users and local government, with a proportion of the committee having to be made up of

women, who are tasked with administration, maintenance and security of markets, but are often not

functional. As part of the market access component, CCRIP helps to organise these groups and

provides them with capacity building support. It also works with the local government to enforce the

legal stipulation that 25 per cent of the market lease income should go to the MMCs for maintenance

costs.

CCRIP is designed to improve the livelihoods of vulnerable women across the IFAD-funded

components, using LCSs as its primary tool. These groups consist of around 25 mainly destitute

women, who are trained and contracted to carry out road and market construction. In selected markets,

Women's Market Sections are also established. These areas are reserved for LCS members and

provide a permanent shop with favourable rent agreements in a safe environment. The project also

provides training to these groups to support other income generating activities. By offering these

opportunities, the project seeks to address the low social and economic status and the skills gap of

women in Bangladesh that restrict their livelihood activities (Roy et al., 2008).

Figure 1 presents the Theory of Change (ToC) for CCRIP, which maps the impact pathways expected

to link the activities of the project through outputs and outcomes to final intended impacts. The ToC

helps to identify the key indicators of success at each stage in order to track the expected impact

pathways of the project (White, 2009). In order for project activities to achieve their intended impacts,

there are a number of contextual factors that are required, which are outlined as part of the ToC.

Outlining these assumed conditions is an important part of the ToC that helps to identify additional

factors that need to be investigated in order to generate a thorough understanding of the project's

impact "story."

10

Figure 1: CCRIP Theory of Change

Women

Form and train Labour

Contracting Societies (LCS)

on construction and other

income generating activities

Contract LCS members to

conduct road and market

construction works.

Roads

Build and upgrade climate

resilient roads, bridges and

culverts

Markets

Build/improve climate

resilient physical markets

and their facilities

Restructure financial

management of markets

and build capacity of Market

Management Committees

Provide information on

farming practices, prices,

and weather through radio

service

Climate change

adaptation

Build/improve cyclone

shelters, upgrade access

tracks

Build community disaster

preparedness cap. building

Improved household

connectivity: schools,

hospitals, financial

services, support

services, etc.

Improved farm

connectivity: input and

output suppliers and

markets, technology/

facilities, other ag.

services such as

livestock vaccination.

Better managed, more

vibrant markets with

more buyers and sellers,

Sustainably structured

market management and

lease payment systems

More climate resilient

road and market

infrastructure

Increased employment

and income generating

capacity of women

Improved capacity to

rehabilitate infrastructure

after shocks

Improved access to

climatic shock protection

Higher education enrollment rates, reduced illness, better social security, etc.

Higher crop productivity and quality from improved input and financial service access

More diverse crop production from improved input and financial service access

Higher volume of goods sold and profits from sale from higher productivity and crop quality, reduced transport costs, value chain inclusion and better prices from better-functioning markets.

Livelihoods less affected by climate stresses and shocks

Diversified household income from improved income generating capacity of women

Increased bargaining power of women due to improved income generating capacity

Increased sustainably and smoothed income

Increased stability and resilience of livelihoods

Increased household and productive asset ownership

Increased food security

Empowerment of women

There is sufficient

demand and institutional

support for the activities

There are no issues with

acquiring land or other

materials for the work

Women are willing and

able to work in LCS

Roads, markets and

shelters are well placed

and well-designed

Training for LCS is

suitable

Farmers face no other barriers to their productivity or their market participation – lack of labour, lack of capital etc.

Income generating capacity is the only barrier to women's empowerment, they face no other barriers.

INPUTS AND ACTIVITIES OUTPUTS

OUTCOMES

IMPACTS

ASSUMPTIONS – Factors that need to be in place for the outputs, outcomes and impacts to be achieved

11

CCRIP is designed to increase household income through a number of intermediate outcomes. First is

increased market participation. With improved roads and accessible, better-managed markets, farmers

are expected to face lower costs associated with bringing their goods to market and thus to sell more.

Selling a higher proportion of their crops should lead to higher incomes, and over time an increase in

household and productive assets. The volume of sales as well as household food security are expected

to also be boosted by higher productivity. With better farm connectivity, and more sellers at markets,

farmers are expected to have better access to productivity-increasing inputs, technology, and extension

services. They may also be able to invest more in their production if improved connectivity leads to

improved access to credit providers.

Another expected intermediate outcome is higher prices. Better market access means more buyers at

markets, more demand, more options, which are likely to drive up prices. Although the upward effect

on prices of increased demand could be cancelled out by a downward effect of increased supply. Before

the project, farmers were often forced into taking lower prices by selling to traders directly after harvest

at the farm gate. With more favourable marketing options and improved access to inputs to improve

crop quality, better market information from CCRIP's radio service, and better connection to post-

harvest processing and storage facilities, this situation is expected to change (Barrett, 2008; Svensson

and Drott, 2010).

In addition to selling more and receiving higher prices, production and marketing expenditures are

expected to decrease, boosting profit margins and adding to the expected income effect. Along with

reduced transport costs, this is expected to occur as improved market access leads to improved input

access, which has the potential to increase the quality and profitability of farmers' crops (Gulati et al.,

2005; Khandker et al., 2009).

The project's work to build climate resilience is expected to ensure the intended outcomes outlined

above are not disrupted by climatic stresses and shocks, and to increase the overall stability and

sustainability of household livelihoods (Meybeck et al., 2012). By making roads and markets more

resistant to cyclones and floods, the vulnerability of livelihoods that are dependent on this infrastructure

is expected to decrease, meaning shocks have lower impacts, and households need less time and

resources to recover after them (Vallejo and Mullan, 2017). The cyclone protection and disaster

preparedness training of KfW and the support to MMCs to increase their capacity to repair markets

after a shock are also expected to contribute to improved climate resilience.

The above effects are targeted at all beneficiaries, whilst the LCS work is designed to produce income

and wellbeing benefits specifically for vulnerable women and women-headed households. In addition to

increased income generating capacity and increased economic opportunities from LCS participation,

LCS members' increased economic independence is expected to lead to the resource and power

allocation shifts needed for increased empowerment and wellbeing (Sheoran, 2016). Whether the LCS

member is the household head or not, members' households are also expected to benefit as LCS

members contribute more to household income and its diversification. Diversification potentially

further boosts the resilience of household livelihoods to shocks (Ellis, 1999; Arslan et al. 2018a).

At the bottom of Figure 1, we identify a number of key assumptions that are required to hold in order

for the above impacts to be fully achieved. If these assumptions do not hold, project's impact could be

constrained. For outputs to be achieved, it is assumed that the project will actually be able to identify

and acquire suitable land for roads, markets and shelters, and that these will be effectively placed, so

that they are used by intended beneficiaries. In terms of translating outputs into outcomes and impacts,

12

the project is designed under the assumption that households face no other significant barriers to their

production or market access such as a lack of capital. Finally, for the LCS activities, it is assumed that

women face no other barriers to their participation, and that women have demand for this support. Once

they have joined, it is also assumed that they face no other economic or social barriers to their

empowerment that the project does not address. The last assumption is a strong one in a country like

Bangladesh, where social norms constrain women in multiple ways, and the implications of this are

discussed further in the results and lessons learned sections below.

c. Research questions

This impact assessment focuses on the impact of IFAD's activities delivered through CCRIP as noted

above. The ToC diagram considers all of CCRIP's components in recognition of the expected overlaps

with the ADB and KfW work. Based on the expected impact pathways of IFAD's activities and the

potential complementarities with the activities of other agencies, this impact assessment answers the

following questions:

1. Did the community roads and markets delivered through CCRIP improve the household and farm

connectivity of beneficiaries? What were the subsequent effects on agricultural productivity, market

participation, and household income?

2. Did the IFAD activities delivered through CCRIP improve the climate resilience of beneficiary

livelihoods? What were the subsequent effects on household income levels and stability?

3. What were the impacts on women's livelihoods from the LCS-related activities? Were there barriers

to their participation in these groups? How effective were the different LCS activities (labour

contracting, income generation training,Women's Market Sections)?

4. What are the contextual factors that may have shaped the impacts of the project on beneficiary

households and women? What other lessons can be learned from the project that can be incorporated

into future rural development, climate resilience, and rural women's empowerment work in

Bangladesh?

13

2. Impact assessment design: Data and methodology

For this impact assessment a combination of quantitative and qualitative data was collected in order

to produce a holistic picture of the project's impact. The main data source is a quantitative household

survey of beneficiary and control households. This section presents the sample design of the

household survey and of the qualitative data collection, along with the statistical methodology

employed for impact analysis, followed by an overview of the key impact indicators used in the

analysis.

a. Overall approach

The CCRIP impact assessment is a collaborative effort between the project team and RIA. In order

for the quantitative household survey to be usable for both this impact assessmet and the project's

Mid-Term Report, we specifically sample beneficiary households who were reached during the early

stages of the project's implementation. In this way, data from these households can be used to

measure the project's performance against the mid-term indicators according to the log-frame in the

Project Design Report. At the same time these households have been exposed to the project's

activities for a sufficient amount of time, the data can also be used to measure the project's overall

impact.

The household survey covered both beneficiary and non-beneficiary households. The key to an

effective impact assessment is to compare a set of beneficiaries (the treatment group) with a set of

non-beneficiaries (the control group), who accurately represent how the set of beneficiaries would

have fared in the absence of the project. In this way, we are able to isolate the effect caused by the

project from other effects that occurred over time. The treatment population of interest for the

CCRIP impact assessment is all smallholder households within the catchment areas of CCRIP roads

and markets. The challenge of this assessment is therefore to identify a representative sample of this

population for the treatment group, and to identify a suitable comparable group of control

households.

To produce the final impact estimates, we conduct an econometric analysis of the household data,

comparing treatment and control households in a model that estimates the size of the effect on each

impact indicator, along with a measure of the effect's statistical significance. Statistical significance

represents the reliability of the result, giving the percentage probability that the result is a reflection

of reality and not due to chance (Gallo, 2016). In order to generate contextual insights, we conduct

our analysis on a range of sub-samples, in addition to the full sample. Table 2 presents the different

sub-samples that we test and the reason for assessing them.

14

Table 2: Overview of sub-sample tests

Sub-sample Hypotheses Tested

Households within 1km of market Did the impact differ based on market proximity?

Households within 2km of market but not 1km of

connecting road vs. Households within 2km of

market and 1km of connecting road

Did being close the connecting road as well as the

market provide a larger impact?

District Did district-specific local factors influence impact?

Households involved in farming activities (crop or

fish production or livestock rearing) vs.

Households not involved in farming activities.

Did project benefits on farm households also spread to

households primarily involved in wage labour,

household enterprises and other off-farm activities?

The type of impact assessment conducted in this report is termed "ex-post" as we use one round of

data collection without a baseline. Ideally the control group would be constructed at baseline, but

without this benefit we employ statistical matching techniques and expert consultations to improve

the accuracy of the treatment and control group comparison. This method has been shown to be

almost as effective as baseline control group construction in some cases (Dehejia and Wahba, 1999).

These matching techniques consist of a variety of matching algorithms to ensure that only similar

households are compared across the treatment and control groups, and is the primary method used

for ensuring accurate impact estimates in the absence of suitable baseline data (Austin, 2011).

b. Data

i. Sample distribution

The sample for the household survey is drawn from eight of the 12 project districts. Given the size of

the area, to collect data from all 12 districts would have been unfeasible and inefficient, thus we

selected eight districts covering the three project divisions (Barisal, Khulna, and Dhaka) according to

those with the largest CCRIP presence and those with the largest number of potential treatment and

control markets. In discussion with the project team, it was decided that the total sample size of the

household survey would be 3,000: a sample size deemed to provide sufficient power to detect

impact, and to cover a large enough area so that the estimation of impact is reliable and

representative.

In terms of the distribution of the sample, the sample frame was designed to achieve

representativeness at the divisional level using data on CCRIP investment by division as a proxy for

the number of beneficiaries in each division. Within each division, the sample is evenly distributed

across the eight districts. Table 3 presents the distribution of the sample across divisions and

districts, based on a sample size of 3,000, with 45 per cent allocated to treatment and 55 per cent to

control.

15

Table 3: Sample distribution across project divisions and districts

Division

Sample

allocation

(%)

Nr.

treatment

households

Nr. control

households

Nr households per

district

Barisal 61 824 1,006 Treatment = 206;

Control = 252

Khulna 16 216 264 Treatment = 108;

Control = 132

Dhaka 23 311 380 Treatment = 156;

Control = 190

The quantitative household data are complemented with qualitative data to contextualise findings

based on input from key informants and focus group participants. The qualitative data therefore

focuses upon the underlying impact mechanisms and the barriers to impact that may have been

faced. The collection of qualitative data was conducted simulatenously with the household survey

and consisted of the following:

Treatment markets: 5 x Key Informant Interviews (KII) with MMC members; 5 x Focus

Group Discussions (FGD) with road construction LCS members; 5 x FGD with market

construction LCS members.

Control markets: 3 x KII with MMC members or Market Manager; 2 x KII with Union

Parishad Womens' Representative.

Project staff: 3 x KII with senior project staff; 3 x KII with regional staff (one for each

project division).

ii. Identification of treatment and control groups

For both the treatment and control groups, we first identified suitable treatment and control markets,

with the intention of sampling households from villages within the catchment areas of these markets.

In order to capture the impact of being close to a CCRIP market, and the incremental impact of being

also close to a CCRIP connecting road, two catchment areas were defined for each market. To

capture the impact of being close to a CCRIP market without a CCRIP road, we sampled households

who are located within 2km of a CCRIP market (or control market) but not within 1km of a CCRIP

connecting road (or a un-improved connecting road in the case of control markets). To capture the

impact of being close to both a CCRIP market and a CCRIP connecting road, we sampled

households who are located within 2km of a CCRIP market and within 1km of a CCRIP connecting

road.

The radius size used for these catchment areas was decided with assistance from the project team.

They explained that these were distances within which households would travel to the market or

road, meaning these were the areas of expected impact. The 2km radius for the markets was also

used for a baseline study conducted of CCRIP. A small number of the control markets did not have a

main connecting road, in which case we sampled all households for the first catchment group.

16

Figure 2 maps the two catchment areas for one of the markets included in the sample (Pazakhali

Bazar). The orange circle represents the 2km radius around the market. The blue zone represents the

catchment area for household located withing 2km of the market and within 1km of the connecting

road. The other catchment area-for households located within 2km of the market but not 1km of the

road-consists of all other areas within the orange circle but not within the blue zone.

Figure 2: Diagram of the two sample catchment areas

Source: Authors' elaboration with technical support from IFAD-ICT Solutions Team.

The treatment markets for the sample were selected from CCRIP markets where market construction

work was completed before July 2015. The period 2013-2015 was considered as Phase 1 of CCRIP's

implementation, and taking households from the earliest phase means they will have been exposed to

CCRIP's work for a sufficient amount of time for impact to develop. Using households linked to

markets where work was completed later risks underestimating project impact by not allowing

enough time to pass before collecting the impact assessment data.

The control markets for the sample were selected from markets on CCRIP's back-up list, along with

additional markets identified by local project staff within the eight project districts covered by the

sample. The back-up list contains 19 markets located in the 8 project districts that were identified for

inclusion in the project according to the project's market selection criteria outlined in Section 1a,

however due to financial reasons these markets were eventually not covered. This means that

households linked to these markets should be suitably comparable to treatment households, and also

facilitates the availability of sufficient information and connections to identify a suitable sample.

Due to the limited number of markets on this list, a further 32 additional control markets were

identified based on expert consultations with the local project staff, who were instructed to identify

markets that would also have been eligible for inclusion in the project at the baseline stage.

Using the above strategies, we identified a total of 46 potential treatment markets for the sample, and

51 potential control markets. In order to identify the most suitable control markets we conducted a

scoring exercise, whereby we asked local project staff to assign scores for each market for a set of

Connecting

road

2km

1km

17

criteria that represented the selection criteria for CCRIP markets1. In this way we were able to build

a better picture of which markets would provide the most accurate counterfactual to treatment

markets. Through this process we eliminated 32 of the potential control markets which were deemed

as unsuitable for inclusion. This left us with 19 potential control markets.

In order to identify the final set of markets that were the most well-matched, we used GIS mapping

to obtain the population densities of the 2km catchment areas around each of the shortlisted

treatment and control markets. This allowed us, first, to identify whether any of the markets had

overlapping catchment areas, and to thus eliminate treatment and control markets that were

overlapping. Secondly, it provided an additional characteristic upon which to match treatment and

control markets—a characteristic that is linked to market size and the income level and

environmental quality of the surrounding area (Cropper and Griffiths, 1994). We therefore selected

the final set of treatment and control markets by producing matched pairs according to population

density and also, when possible, from within the same upazila to avoid the between upazila socio-

ecological hetorogeneity.

For each treatment and control market selected for the sample, the distribution of the sample across

the two catchment groups is equal in order to obtain a sufficient sample size to measure impact on

both of the catchment groups. From within the catchment areas of each market, households were

selected randomly. As we did not have lists of all households within each area from which to

conduct a purely random selection, we devised a three-stage process. Firstly, for each market, with

the assistance of the local leaders (primarily the Union Parishad Chairman), we compiled lists of all

of the villages located in each of the catchment areas, along with an estimate of the number of

households in each village. Our target was to collect data from a minimum of 15 households per

village, meaning the second stage involved randomly selecting the villages to be included in the

sample in order to meet this target. In cases where there was an insufficient number of villages in the

catchment area to meet the target, we included all villages located in the catchment area and

increased the sample size per village. For the final stage, within each village, households were then

randomly selected through the random walk method, using a sampling interval based on the

estimated village population size.2

Based on this strategy, the final set of treatment and control markets, and their associated sample

sizes are presented in Table 4 below, and Figure 3 contains a map of the sampled area where the

sampled markets and their catchment areas are plotted.

1 To reflect the CCRIP market eligibility criteria, scores were assigned for the following: (i) Based in char, low-lying,

remote, disaster-prone and infrastructure poor area (Yes/Somewhat/No) ; (ii) Connecting roads to market are dirt roads

that are not flood resistant (Yes/Somewhat/No); (iii) Market has a multi-purpose shed (Yes/No); (iv) Market has a fish

shed (Yes/No); (v) Market has a boat landing platform (Yes/No); (vi) Market has an open paved/raised area (Yes/No);

(vii) Market has a women's section (Yes/No); (viii) Market has an internal road (Yes/No); (ix) Market has improved

drainage (Yes/No). 2 For this method we first selected a landmark within the village (such as a school or communal area) and assigned four

teams to walk north, south, east and west of the landmark. The teams were assigned to sample households in their

direction according to a pre-assigned interval. This interval was calculated based on the total population of the village

and the required sample size for the village. We divided the total population by four (based on the four directions), and

the required sample size by four (based on the four directions and four teams), and then calculated the interval by

dividing the share of the total population by the share of the sample size. For a village with 100 households and a

required sample size of 16, the calculation would be as follows: (100/4)/(16/4) = 25/4 = 6.25. Meaning the sampling

interval would be every 6 households in a given direction.

18

Table 4: Distribution of final sample

Market name Division District Upazila Treatment

group

Sample

size

Joyer Hat

Barisal

Bhola

Burhanuddin

Treatment 103

Chanmiar Hat Control 126

Uttar Manika Bazar

Charfession

Treatment 103

Dolar Hat Control 126

Pazakhali Bazer

Patuakhali Sadar

Treatment 206

Akhai Bari Control 252

Gutia Bazar

Barisal Uzirpur

Treatment 206

Gondershor bazar Control 252

Badurtala Hat

Barguna

Pathorghata Treatment 206

Barotaleshwar Bazar Bamna Control 252

Chutukar Hat

Khulna

Bagerhat Sharankhola

Treatment 108

Rajapur Bazar Control 132

Ghorkumarpur Bazar

Satkhira Shyamnagar

Treatment 108

Patakhali Bazar Control 132

Suagram hat

Dhaka

Gopalgonj Kotalipara

Treatment 156

Hasua Bazar Control 190

Bairagir Bazar

Madaripur Rajoir

Treatment 156

Sonapara Bazar Control 190

Total 3,004

19

Figure 3: Map of treatment and control markets

Source: Authors' elaboration with technical support from IFAD-ICT Solutions Team. Each market is

represented by its centroid on the map and the circles around the centroid represent the cathment

area within 2km of each market.

Table 5 presents an overview of key descriptive statistics of the treatment and control groups, giving

an idea of the effectiveness of our sampling strategy in constructing comparable treatment and

control groups. Although in most cases the statistics are reasonably similar across the two groups,

there remains important statistically significant differences in household composition and religion,

and the average distance of the household to the sample market. This highlights the challenge of

creating accurate comparison groups ex-post, and underlines the need to use additional statistical

techniques as in this impact estimation analysis to minimise these remaining differences.

20

Table 5: Comparison of treatment and control samples

Treatment

(1,188 households)

Control

(1,552 households) Difference

Household size 4.57 4.53 0.04

Dependency ratio (%) 61.66 69.27 -7.61***

Age of h'hold head 49.82 48.31 1.51***

Average household age 31.56 30.51 1.05**

Religion (%):

Muslim

Hindu

Christian

83.75

14.81

1.43

94.85

5.15

0

-11.10***

9.66***

1.43***

Gender of household head (%):

Male

Female

90.15

9.85

86.21

13.79

3.94***

H'hold head is literate (%) 76.09 73.32 2.77*

Nr. literate h'hold members 2.02 1.92 0.10**

Nr. climatic shocks in the past year 0.15 0.16 -0.01

Distance to sample market (km) 1.21 1.09 0.12***

Note: *,** and *** indicate that results of a t-test for the difference between the treatment and

control groups, representing whether the difference is 90, 95 and 99% significant, respectively.

c. Questionnaire and impact indicators

The household questionnaire collected a wide range of information to create the impact indicators

and other variables used in the data analysis. The questionnaire was designed to collect detailed

information on agricultural production (separated by crop and by the three main cropping seasons),

fish production (separated by pond), livestock rearing, income from other sources, asset ownership,

food consumption, financial inclusion, shock exposure, social capital, access to services, and

household decision making. The questionnaire was conducted between late August and early

November 2018 and collected information covering the 12 month period between August 2017 and

July 2018.

In addition to the qualitative and quantitative data, the project team have also conducted the

following additional surveys: (i) Survey of LCS members; (ii) Child anthropometric measurement;

and (iii) Surveys of road and market performance, some of which are used to inform this study.

Based on the ToC for CCRIP presented in Figure 1, we use the statistical analysis outlined in the

proceeding section to test the project's impact on a wide set of indicators. The agricultural indicators

and those relating to fish production are all aggregated across plots and ponds to the household level.

21

All of the indicators used, along with how they are constructed and the impact areas they are linked

to are presented in Table 6.3

3 Appendix I contains the mean values for the treatment and control groups for all of these impact indicators.

22

Table 6: List of impact indicators

Indicator Description Impact area

Agricultural and fish production and sale

Gross value of agricultural/fish

production

Converts harvest of all crops/fish into a monetary unit. Equal to the income from crop/fish sales, plus the

value of non-sale uses, valued using the median price for the sample for each crop/fish when sold (Carletto

et al., 2007).

Volume of production; effectiveness/efficiency

of farming/fishing practices.

Gross margins/ha Equal to the gross value of production minus the value of all inputs. Inputs that were not purchased were

valued using the median price for the sample for each input. Effectiveness/efficiency of farming practices.

Proportion of harvest sold Gross value of the crops sold as a percentage of the total gross value of crop production. Market participation and access

Sold at a market Household sold at least a portion of its crop/fish harvest at a market, rather than from home or farm gate Market participation and access

Total revenue from crop/fish sales Cash income received from sale of all crops/fish. Market participation and access; income.

Nr crop varieites Count of the different crops grown. Input access, farming practices, shock

resilience.

Grew high value crops Household cultivated at least one high value seasonal or perennial crop Input access, farming practices

Land cultivated Number of hectares of land cultivated. Input access, wealth

Value of inputs (BDT/ha.) Calculated for agriculture and fishing. Total monetary value of all inputs used. Inputs that were not

purchased were valued using the median price for the sample for each input.

Input access; investment in agriculture;

effectiveness/efficiency of farming/fishing

practices.

Productivity of inputs (BDT) The gross value of production per one Taka of input used. Input access; effectiveness/efficiency of

farming practices.

Value of fish consumption

(BDT/capita) The gross value of fish that was produced and used for home consumption Market access and participation; nutrition

23

Table 6: List of impact indicators (cont'd)

Indicator Description Impact area

Livestock ownership and production

Gross value of livestock production Calculated in the same was as for crops and fish. Median price used to value livestock and

livestock products that were consumed at home or given away as gifts.

Livelihood practices, effectiveness/efficiency

of livestock prod.

Net value of livestock production Calculated in the same way as for crops and fish Livelihood practices, effectiveness/efficiency

of livestock prod.

Income from livestock activities Cash income from sale of whole livestock and livestock products, including cuts of meat, milk,

eggs and manure.

Effectiveness/efficiency of livestock prod.,

income.

Milk productivity (ltr/cow) Amount of milk produced per adult cow Input access; Effectiveness/efficiency of

livestock prod.

Livelihood composition and poverty

Gross household income per capita (BDT) Total cash income from all sources per household member Income, poverty.

Net income per capita (BDT) Equal to gross household income minus all economic expenditures Income, investment, livelihood effectiveness.

Above poverty line Household income is above the $1.90 per person per day poverty line set by the World Bank

(Ferreira et al., 2015) Poverty

Number of income sources Count of the number of different sources of income in the household Livelihood practises; livelihood resilience.

Income compositon Percentage of household income comprised from crop sales, livestock, formal and casual wage

labour, household enterprise, remittances, land rental and other (interest, pension, etc.) Livelihood practises, income

Sector participation A yes/no indicator for household being involved in the following sectors to earn income:

agriculture, fishing, livestock, waged labour, and household enterprise Livelihood practises

24

Table 6: List of impact indicators (cont'd)

Indicator Description Impact area

Assets, food security and education

Asset indices Composite indices constructed using Principle Component Analysis, calculated separately for

Productive assets and household assets (Filmer and Pritchett, 2001) Wealth, shock resilience.

Tropical Livestock Units Count of livestock owned, with different livestock converted into common unit (Jahnke, 1982). Wealth, shock resilience.

Food insecurity experience Standard indicators of food insecurity also adopted by SDGs (2.1.2), responses to eight questions

tallied into one Food Insecurity Experience Score (FIES) (Ballard et al., 2013). Wealth, food security

Dietary Diversity Score Score based on the consumption of different food groups in the past seven days (FAO, 2010). Nutrition

Childrens' school enrolment Proportion of school-age children currently enrolled in school Education

Financial inclusion

Have formal bank account or other

account

Yes/no indicators for whether the household has a formal bank account (held with a registered bank),

and whether they have another type of account, such as those held with a microfinance institution, a

non-bank financial institution, or a mobile money account (Anderson et al., 2016)

Wealth, market access

Received loan Yes/no indicator of whether household has received at least one loan in the past year Market access

Cash savings Total cash savings per capita, totalled across all savings locations. Market access, shock resilience, income

Women's empowerment

Womens' autonomous income

generation

Proportion of household income from the wage labour of female household memebrs or from

household enterprises owned or managed by female household members. Women's empowerment

Womens' decisionmaking involvment Female household members are involved (either individually or jointly) in decisions regarding:

household purchases, children's education, farm and livestock production and sale. Women's empowerment

25

d. Impact estimation

The main aim when estimating impact is to ensure that there is minimal difference between the

treatment and control group that are being compared. As well as our efforts to ensure that the two

groups are comparable during the sampling stage, as outlined above, we also employ statistical

techniques to further improve comparability during the data analysis stage. There are a number of

different statistical techniques that can be used to improve comparability and we use two separate

analytical models to test CCRIP's impact. Given the variety of approaches to ensuring comparability,

it is best-practice to employ one primary approach and one secondary approach that serves as a

means of testing the robustness of the results from the primary model. If the results are qualitatively

similar across the two approaches, then we can be assured that the results are valid and are not

model-specific.

Before running our impact estimation models, data quality checks were performed and 142 treatment

and 90 control households were removed from the final sample, either because the household was

incorrectly sampled from outside the 2km sampling zone around the focal market, or because their

data on key impact indicators (crop or fish harvest and income) were identified as outliers. In cases

where households had outlier data for minor variables, these values were replaced with imputed

values based on the distribution of the rest of the data.

Using the remaining sample, we then conducted an initial round of trimming using Propensity

Scores, whereby households that were clearly very different from the rest of the sample were

identified and dropped as they were unlikely to have comparable households in the opposite group

(Caliendo and Kopeinig, 2008). The propensity scores represent the likelihood of a household being

selected for project inclusion based on a set of relevant pre-project variables (or variables that are

unlikely to have been affected by the project), and are commonly used in impact assessment to

identify treatment and control units that would have been similar before project implementation, in

the absence of baseline data.4 Similar scores between a pair of treatment and control households

suggests that the two households were in a similar situation before the project began, meaning that

the control household can provide an accurate picture of how the treatment household would have

fared in the absence of the CCRIP.

Once we created these scores, we identified and dropped all households that had a Propensity Score

outside the area of common support, meaning that we dropped households with a score that was

above the highest score from the opposite group, and all those with a score below the lowest score in

the opposite group. For treatment households, the lowest score was 0.18 and the highest score was

0.88, and for control households, the lowest score was 0.17 and the highest score was 0.87. Based on

this we dropped ten treatment households and three control households.

In addition, we also dropped all treatment and control households that did not have at least one

household in the opposite group whose score was within 0.01 of their own score. From this, we

dropped an additional 13 treatment and 17 control households (see Appendix II for the distribution

of the propensity scores before and after these households were removed).

4 Specifically, the Propensity Scores in this case were created by running a logistic regression model where the dependent variable

is the binary treatment status (treatment or control) and the independent variables are variables linked to CCRIP selection and/or

livelihood capacity from 2014 (i.e. before the project was implemented), and then deriving the scores (representing the probability

of treatment) using the coefficients for each of the independent variables in the model, which represent their effect on the likelihood

of being treated (Caliendo and Kopeinig, 2008).

26

With the trimmed datset that consists of 1,192 treatment and 1,546 control households, we then run

our analytical models. The primary model we employ is a nearest neighbour matching (NN) model

(Khandker et al., 2010; Austin, 2011). This is similar to the approach used in the trimming process,

in that it creates Propensity Scores using the same set of matching variables, and identifies all

matched pairs within a set radius, although we narrow the radius in the model to 0.001 to increase

precision.5

The radius we have set for this model is relatively small, meaning that only the most well-matched

households are paired in the model. It is not always possible to set such a small radius size, however

because our treatment and control groups were already similar due to the efforts taken in the

sampling stage (as shown in Table 5), we were able to set this small radius with the confidence that

there are sufficient matches within the radius, and that we would not have to drop households that

did not have any matches within the radius.

One of the main benefits of the NN model is that it allows us to ensure that households are matched

exactly on variables of particular importance. As noted, we split the sample into two catchment areas

for each market: those 2km from the market but not from the connecting road, and those 2km from

the market and 1km from the connecting road. In order to ensure that we are not comparing

households located in different catchment areas, we implemented exact matching within each

catchment area as part of the NN model. Using GIS data, we also investigated implementing exact

matching on the distance of the household from the market (e.g above or below the median distance),

but we found that the level of comparability was best achieved when this distance was included in

the set of matching variables used to create the Propensity Scores, rather than using a categorised

version of this variable in exact matching.

Once the matched pairs are created using the NN model, impact is then estimated by taking the

average of the difference between the matched pairs. For example, if we compare the income per

capita of the treatment household with that of the control household within all of the matched pairs,

the impact estimation, termed as the Average Treatment Effect (ATE), is the average of the

differences in income across the pairs. The ATE is defined as:

Average Treatment Effect = E(y1i – y0i) (1)

Where y1i represents the outcome for treatment household i, and y0i represents the outcome for

control household i, and the E is the expectations operator

The secondary model we employ is an Inverse Probability Weighted Regression Adjustment

(IPWRA) model (Wooldrige, 2010; Austin and Stuart, 2015). In this model, impact is estimated

using a weighting rather than a matching approach. The model first assigns a weight to each

household in the analysis that represents the inverse of the probability of their receiving the

treatment that they actually received (beneficiary or control), with the probability calculated based

on the same set of variables used to create the Propensity Scores in the NN model. An econometric

regression model is then run with the weights applied, which estimates the average expected value of

the outcome if all units in the sample received the project and if all units in the sample did not,

5 A number of different radius sizes for this model were tested, and this radius produced the strongest set of

comparable matches whilst minimising the need to drop households who do not have a match (which is the downside

to using smaller radius sizes).

27

controlling for a set of relevant covariates, with the final impact estimate being calculated by

subtracting the control outcome estimate from the treatment estimate.

The set of control variables used in the IPWRA model cover household socio-economic

characteristics, land characteristics (for the agriculture indicators), geographic area, and exogenous

shock exposure. Control variables were chosen based on their likelihoods to have influenced the

outcome variable, while not having been affected by the project; thus different sets of control

variables were used depending on the outcome variable being analysed 6.

As mentioned, there is a wide variety of potential analytical models to employ for impact estimation,

and these must be selected based on which model provides the most reliable comparison between

treatment and control. In this case we tested a number of different models and found that the NN

model and IPWRA model were the most effective in this instance. The main way that we assessed

the effectiveness of the models was by assessing the change in the Standardised Mean Difference

(SMD) and in the Variance Ratio (VR) of the set of matching/weighting variables used for the

analysis (Austin, 2009).

The SMD measures the difference in the treatment and control means of a variable by the difference

in standard deviations, allowing for differences across variables to be compared in the same unit.

The VRs give a picture of how the relative variation across the treatment and control groups for each

matching variable has been altered. Table 7 contains the set of variables that were used in the

matching for the NN model and the weighting for the IPWRA model, and presents the change in the

SMD and the VR for each variable before and after the models are applied.

The average SMD between the treatment and control groups across the variables in the unadjusted

sample is 9.5 per cent, and we see that this is reduced to 2.1 per cent by the NN model and to 2.5 per

cent by the IPWRA model. For the average VR, we see that this has been reduced from 1.2 for the

raw sample, to 1.1 by the NN mode and to 1.1 by the IPWRA model. These statistics show that we

were able to significantly improve the comparability of the treatment and control groups for the

impact estimation with both models, highlighting that we can interpret our results with confidence

that there is negligible bias in the comparison. As the NN model performs slightly better in these

tests in addition to allowing for exact matching on catchment area, we employ the NN model as the

primary model for our analysis.

6 See Arslan et al. (2018b) for a detailed statistical specification of this model.

28

Table 7: Change in SMD and VR from the two impact estimation models

dd Standardised Mean Difference Variance Ratio

Variables Raw

(%)

After

NN (%)

After

IPWRA

(%)

Raw After

NN

After

IPWRA

HH size 1.83 -3.72 1.29 1.23 1.06 1.15

Dependency ratio -13.45 -3.13 2.64 0.73 0.84 0.89

Age of HHH 9.96 2.04 -2.82 0.99 0.97 0.98

Gender of HHH -11.66 0.45 1.50 0.76 1.01 1.04

Age * Ed of HHH 8.83 1.86 -4.19 0.94 0.95 0.88

Mean HH age 9.06 3.14 -1.86 0.99 0.97 0.96

Ed of HHH 7.00 1.28 -3.38 0.92 0.95 0.92

Ed*Gender of HHH 2.91 0.56 -2.64 0.82 0.91 0.92

Nr literate in HH 2.07 0.52 3.02 1.06 1.02 1.01

Nr English speakers in HH -0.84 -3.24 2.86 1.21 1.05 1.17

HHH speaks English -2.33 -0.26 1.08 1.07 1.01 0.97

Muslim -36.56 -4.39 1.77 2.78 1.13 0.97

Nr climate shocks -1.66 -0.91 -2.24 1.07 1.02 1.05

Productive asset index 2014 9.04 -4.48 1.78 1.92 1.33 1.86

Household asset index 2014 2.33 -2.46 3.26 1.12 0.96 1.16

Catchment area -28.01 0.00 0.65 1.20 1.00 1.00

Upazila 4.73 1.46 5.00 0.99 1.03 1.16

Distance to market 18.02 4.08 -3.42 2.44 1.66 2.40

Average (in absolute terms) 9.46 2.11 2.52 1.24 1.05 1.14

† These are interaction terms whereby two variables are multiplied together to incorporate their combined

effect (Baser, 2006).

29

3. Profile of the household questionnaire sample

a. Household characteristics by division

CCRIP is spread across three of Bangladesh's seven divisions: Barisal, Dhaka and Khulna. There is a

high level of diversity across these divisions that is likely to have an influence on the impact of the

project (Ahmed et al., 2013). For instance, as can be seen from Figure 3, the districts sampled for

Barisal and Khulna are located along the coastal areas and char zones, whilst the districts sampled

for Dhaka are located further inland. As noted, households living in the char zones face specific

challenges to their livelihoods, most salient of which is the more intense seasonal flooding in the

area, as well as being more remote (Sarker et al., 2003). There is also disparity in wealth, with Dhaka

the wealthiest of the three divisions, followed by Khulna (Ahmed et al., 2013).

Table 8 presents socio-economic statistics of the household sample for the three project divisions.

The divisions are similar in terms of average household size (around 4.5 members), average age of

the household head (around 49) and average age of household members (around 31). The gender of

the household head and the religion of households are similar in the Barisal and Khulna divisions—

with between 90-94 per cent of households having a male head, and between 93-96 percen belonging

to the Muslim faith, and the remainder belonging to the Hindu faith. Households in Dhaka district

are quite different in these characteristics, however, with a higher proportion of female-headed

households (around 23 per cent), a lower proportion of Muslims (76 per cent) and more Hindus (21

per cent) and Christians (3 per cent) compared to the other divisions. It is well-documented that

female-headed households face specific livelihood barriers, including higher poverty, lower access to

productive assets and capital, and constraining social norms, thus, we interpret the impact findings

with this background in mind (World Bank, 2001).

The households sampled from Dhaka are the wealthiest, whilst Barisal and Dhaka have similar,

lower levels of income. This wealth distribution is also reflected in the poverty levels of the three

divisions, with 82 per cent of households above the income poverty threshold in Dhaka, compared to

71 per cent in Barisal and 76 per cent in Khulna. The national poverty rate is reported to be around

15 per cent in the country,7 which indicates that CCRIP has effectively targeted areas with higher

levels of poverty and lower average incomes compared to the national average of BDT127,255 per

capita (around US$1,6048) (World Bank, 2017).

The project's targeting of households most exposed to climatic shock risk is also reflected in the

livelihood distribution statistics. In none of the divisions does a single income source provide more

than a third of total income, with households in all three regions showing signs of livelihood

diversification, which is a common form of ex-ante risk management in the face of persistent shock

threats (Jones et al., 2010).

Although the households sampled from the three divisions all have diversified livelihoods, there is

variation in how they diversify. In the Barisal division, around 18 per cent of household income is

7 http://povertydata.worldbank.org/poverty/country/BGD 8 Conversion made on 31 January 2019 from the following source:

https://www.reuters.com/finance/currencies/quote?srcAmt=127255&srcCurr=BDT&destAmt=&destCurr=USD

30

derived from the sale of crops, whilst 24 per cent of households do not cultivate any crops at all. In

Dhaka, crop sales provide 13 per cent of income and 34 per cent of households do not cultivate any

crops. In Khulna, just 5 per cent of income is from crop sales and 53 per cent of households do not

cultivate any crops. In this division, fish and livestock production play a more prominent role.

Across the divisions, formal wage labour and household enterprises are the major sources of income

for the households in the sample, although casual wage labour provides around 33 per cent of

income for households in Khulna, as expected given that housheolds in this division have the lowest

average income per capita. Household enterprises provide around 20 per cent of income in Barisal

and Khulna, and more than 30 per cent in Dhaka. The prominent role played by remittances is

evident everywhere, providing at least 10 per cent of total income, and they are especially important

in Dhaka providing around 27 per cent of household income, most likely coming mainly from those

working in the capital city. From these statistics it is clear that our impact analysis must focus on all

aspects of households' livelihoods, rather than simply agriculture, with improved access to waged

labour opportunities and demand for household enterprise services being key potential impact

mechanisms for CCRIP.

Although the contribution of wage employment to household income is high in all regions, there is

variation in the types of employment. In Barisal, only nine per cent of the formal employment is on

the farm, whilst 53 per cent is professional work (most commonly either office work or technical

jobs such as a mechanic), and 37 per cent is other off-farm work (such as construction). In Khulna,

professional work accounts for 86 per cent of formal employment, with nine percent constituted by

other off-farm work, and just five percent on-farm work. Finally for Dhaka, the largest share of

formal employment is provided by other off-farm work (44 per cent), with on-farm waged labour

accounting for a much larger share compared to the other regions (30 per cent).

Regarding household enterprises, in all divisions the majority (around 30-40 per cent) consist of non-

agricultural businesses such as small shops or services such as carpentry or barbering. In all districts,

taxi or pick-up services account for around 20 per cent of businesses, and in Dhaka, enterprises

offering professional services such as midwifery or tutoring are also common (around 30 per cent).

31

Table 8: Socio-demographic and livelihood characterisitcs of household sample by division

-1-

Barisal

-2-

Dhaka

-3-

Khulna

Household characteristics

Household size 4.56 4.55 4.46

Age of h'hold head 48.97 48.56 49.48

Average household age 31.08 30.37 31.30

Household dependency ratio (%) 63.27 77.65 60.75

Religion (%):

Muslim

Hindu

Christian

93.26

6.74

0.00

76.43

20.71

2.86

95.81

4.19

0.00

Gender of household head (%):

Male

Female

90.01

9.99

76.60

23.40

94.93

5.07

Per cent of adults who are literate

(%) 65.27 68.54 72.06

Per cent of school-age children in

school (%) 70.20 (1,408) 69.46 (505) 72.17 (371)

Livelihood characteristics

Total income per capita (BDT) 69,307.87 95,006.85 62,135.35

Above $1.90 per day poverty line

(%) 70.62 82.66 76.43

Land owned with title (ha.) 0.26 0.33 0.16

Proportion of income from (%):

Agriculture

Fish

Livestock

Formal wage labour

Casual wage labour

Household enterprise

Land rental

Remittances

Other (pension, etc)

17.60

1.30

9.09

24.76

8.37

21.22

1.03

11.79

4.29

13.40

0.67

5.64

14.06

1.92

30.25

2.25

27.40

4.38

5.24

4.17

4.30

17.43

32.98

22.78

0.52

9.03

3.08

Note: Unless otherwise stated in parentheses, the numbers of observations for these values are as

follows: Barisal = 1,692; Dhaka = 594; Khulna = 454.

32

Table 9 presents statistics on the agricultural practices of the households in the sample by division.

The more prominent role played by crop production in Barisal is reflected in the much larger harvest

value compared to the other two regions, both in terms of the volume produced, and the productivity

of production per hectare. Despite this, the largest average income from crop sales is for households

in Dhaka, followed by Barisal and Khulna, which is in line with the average total incomes across the

divisions.

The statistics for the dry and monsoon seasons relate only to the production of seasonal crops. In the

case of Barisal and Khulna, a larger proportion of households cultivated seasonal crops in the

monsoon season than in the dry season. Conversely, in Dhaka a larger proportion of households

cultivated seasonal crops in the dry season. This variation could potentially be linked to the types of

crops produced in Dhaka, as well as differences in seasonal demand for other income generating

activities (wage labor and household enterprises, both of which are more important in this division).

The importance of perennial crops is highlighted for Barisal in that a much smaller proportion of

crop production (in terms of value) is accounted for by seasonal crops, whilst seasonal crops provide

the main contribution in Dhaka and Khulna. The climate-related production issues in the project area

are reflected by the seasonal productivity statistics for the divisions, with total productivity higher in

all cases for the dry seasons, even in Dhaka where total production volume is higher during the dry

seasons.

In terms of income from crop sales, there are important differences in the selling practices of the

three divisions. In Barisal and Dhaka, around 70 per cent of crop-producing households sold some of

their harvest, while this figure is 50 per cent in Khulna. In terms of seasonal crops, in Barisal and

Dhaka the income per hectare from selling these crops is higher in the monsoon season. In total,

households in Dhaka sold the highest proportion of their crops (37 per cent), compared to 29 per cent

in Barisal and 22 per cent in Khulna. As is common for smallholders across the country, the majority

of households' harvest is dedicated to home consumption in all divisions (Anderson et al., 2016).

However, this is most prominent in Khulna (69 per cent of harvest), likely linked to the higher

poverty level in this division and also the higher production of rice, which is more likely to be

consumed at home than other crops (ibid). Households in Barisal lose the highest proportion of their

harvest to pests and shocks, although this percentage is low in all cases.

In terms of the types of crops that are grown, the main crop in all three divisions is rice, which is

grown by between 72 per cent and 85 per cent of the sample in the three divisions. In Barisal, beans

are an important crop, although not so in the other two divisions. Other divisional differences are the

importance of khesari in Khulna, and of coconuts and betel nut in Barisal.

33

Table 9: Agricultural production and marketing practices of household sample by division9

-1-

Barisal

-2-

Dhaka

-3-

Khulna

Total value of harvest (BDT)

Annual (incl. perennials)

Dry seasons

Monsoon season

933,839.50 (1,274)

23,508.84 (807)

22,188.45 (937)

54,899.21 (390)

52,820.32 (336)

18,129.56 (139)

34,478.86 (212)

14,707.32 (48)

28,407.84 (188)

Total value of harvest per ha.

(BDT/ha)

Annual (incl. perennials)

Dry seasons

Monsoon season

97,499.74 (1,274)

63,743.00 (807)

60,909.70 (937)

81,867.56 (390)

104,188.80 (336)

48,724.83 (139)

49,568.06 (212)

87,722.70 (48)

59,631.42 (188)

Total income from crop sales

(BDT)

Annual (incl. perennials)

Dry seasons

Monsoon season

60,073.53 (890)

18,904.40 (616)

27,266.45 (667)

75,782.55 (272)

39,581.44 (250)

51,603.67 (137)

28,104.85 (107)

14,508.33 (12)

27,498.93 (95)

Total income from crop sales per

ha. (BDT/ha)

Annual (incl. perennials)

Dry seasons

Monsoon season

34,071.75 (890)

44,689.18 (616)

62,481.47 (667)

87,590.46 (272)

63,191.68 (250)

163,600.20 (137)

26,418.81 (107)

50,061.43 (12)

38,874.06 (95)

Total area cultivated (ha.)

Annual (incl. perennials)

Dry seasons

Monsoon season

3.30 (1,274)

0.39 (867)

0.40 (937)

1.92 (390)

0.54 (337)

0.41 (142)

3.13 (212)

0.29 (68)

0.53 (188)

Crops grown (% of sample):

Rice

Beans

Khesari

Coconut

Betel nut

75.59

37.91

3.61

19.31

22.45

76.41

1.79

24.87

9.49

3.59

90.57

0.94

0.47

12.74

8.96

Harvest uses (% of total value):

Home consumption

Sold

Lost

Used for feed

Used for seed

Used for other purposes

60.47

28.95

1.19

0.20

1.15

8.04

52.35

36.50

0.73

0.13

1.43

8.86

68.89

22.17

0.37

0.66

1.35

6.56

Note: Unless otherwise stated in parentheses, the numbers of observations for these values are as

follows: Barisal = 1,274; Dhaka = 390; Khulna = 212.

9 Average statistics for annual, and seasonal indicators are compiled only for those households who cultivated during

each period. This is why the number of observation vary across indicators and why the annual average amounts are

not always above the seasonal average amounts.

34

b. Household characteristics by catchment area

Table 10 presents statistics of the sampled households separated by their location within the 2km

radius from the treatment or control market. The first column contains statistics for the full sample,

covering all those within a 2km radius of the market; the second is for all households located within

a 1km radius of the market, the third is for households located within 2km of the market but not

within 1km of the connecting road; and the fourth is for households located within 2km of the

market and also within 1km of the connecting road. For the treatment group, these different

groupings represent different levels of exposure CCRIP's support, with those located closer to the

market (Column 2) and those located closeer to both the market and the connecting road (Column 4)

having the highest level of exposure and thus being expected to receive the highest benefit.

From the table we see that the highest average income is amongst households located closer to the

connecting road, whilst there does not seem to be a difference in income for households according to

their distance from the market (based on statistics for columns 1 and 2). The lowest average income

is for households who are located within 2km of the market but not within 1km from the road,

suggesting that proximity to the connecting road is a key determinant of income variation. This

difference in wealth is not reflected by land ownership, however, with this figure averaging around

0.25 ha. for all groups.

One may expect that households located closer to the road would have higher school enrolment due

to better access, but this averages around 70 per cent in all cases. This may be due to the existence of

local schools in the area that do not require main connecting roads to access. The higher income of

people closest to the road is definitely highlighted, however, by the value of harvest, which is over

double the amount for some of the other groupings. Interestingly for this statistic, the value is not

higher for households located closer to the market.

In terms of income composition, households are similar across the different groupings, suggesting

that there is little variation in livelihood practices according to location in the catchment area. In all

cases, the majority of income is provided by household enterprises, followed by formal wage labour,

with income from crop sale providing around 13-16 per cent in all cases.

35

Table 10: Household and livelihood characteristics of the household sample by catchment area

-1-

2km from

market

-2-

1km from

market

-3-

2km from

market only

-4-

2km from market

and 1km from

road

Total income per capita (BDT) 73,690.67 73,045.28 69,638.25 78,528.26

Land owned with tenure (ha.) 0.26 0.25 0.26 0.26

School-age children in school (%) 70.36 (2,284) 69.59 (1,096) 69.75 (1,248) 71.10 (1,036)

Value of harvest per ha.

(BDT/ha) 88,833.39 (1,876) 48,939.31 (866) 46,025.22 (985) 136,157.80 (891)

Total area cultivated (ha.) 3.00 (1,876) 2.84 (866) 2.83 (985) 3.18 (891)

Proportion of income from (%)

Agriculture

Fish

Livestock

Formal wage labour

Casual wage labour

Household enterprise

Land rental

Remittances

Other (pension, etc)

14.64

1.64

7.55

21.23

11.05

23.44

1.21

14.72

4.11

13.12

1.22

7.34

20.24

11.90

24.62

1.11

16.33

3.83

13.76

1.57

6.54

22.68

12.79

23.01

1.17

13.92

3.98

15.69

1.73

8.76

19.50

8.97

23.94

1.26

15.68

4.26

Note: Unless otherwise stated in parentheses, the number of observations for each column are 2,740;

1,312; 1,491; and 1,249, respectively.

36

4. Results

In this section we present and discuss the results for CCRIP's impact on the set of indicators listed in

Table 6. We first present the results for CCRIP's impact according to households' proximity to the

CCRIP market, presenting results for the full sample from the 2km catchment area along with the

results for households located within 1km of the market. We then present results for CCRIP's impact

according by geographic location; by households' proximity to both the market and the connecting

road; and according to households' livelihood activities. The results we present are from the primary

NN model, whilst the results from the secondary IPWRA model are presented in Appendix III.

It is important to note that for the impact on specific sources of income (such as crop sales, fish

sales, or wage labour), the results apply only to those households, who were participating in these

activities. For instance, the impact we report on income from crop sales refers only to those

households who actually cultivated crops, and that for the impact on wage labour only applies to

those who actually worked as wage labourers during the study period.

a. Overall impacts of CCRIP

i. Agricultural production

Table 11 presents the impact of CCRIP on agricultural productivity and crop sale. For the whole

sample, we do not find a statistically significant impact on agricultural production for the 2km

catchment area for the whole 12 month cropping period. We find significant variation in impact by

season for seasonal crops, however, with a statistically significant increase in crop production of 94

per cent for the dry seasons (Kharif I and Rabi), but a negligible impact in the monsoon season

(Kharif II). In terms of gross margins, which is a measure of production efficiency, the project did

not have a significant impact for the full 2km catchment area. We also do not find that the project

increased the number of crop varieties grown by beneficiary households within the 2km catchment

area.

For the 2km catchment area, the results for crop sales differ from the crop production results. We

find that, for the full 12 month cropping period, income from crop sales increased by 104 per cent,

with a significant impact being achieved both in the dry and monsoon seasons—although the dry

season impact is slightly higher. We find that this impact is achieved despite the lack of a large

signficiant impact on agricultural production, partially due to a larger proportion of harvests being

sold amongst beneficiaries rather than being used for home consumption or lost to pests. In addition,

we find that beneficiary households were 11 per cent more likely to have sold their crops at a market,

rather than from home or at the farm gate, and were eight per cent more likely to have grown high-

value crops, impacts which also likely contributed to the total income effect.

For the 1km catchment area, there is evidence in some cases of an intensification of impact in terms

of crop production and sales. Although the impact on the value of crop harvest is still not significant,

we find a significant increase in gross margins of nine per cent, suggesting that production efficiency

was improved more for households located closer to the CCRIP markets. The effect on crop sale is

108 per cent for households located within the 1km radius as compared to the 104 per cent for those

in 2km radius, although the impact on seasonal crop income is not significant for the dry or monsoon

37

seasons, suggesting that impact on crop income came mainly through perennial crops for these

households.

The main implication of these findings is that improved market access for the sale of crops has

clearly been achieved through CCRIP, and has contributed to a very large increase in on-farm

income as a result. The statistically signficiant impact on crop sale income for seasonal crops in the

monsoon season is particularly salient with regards to this project, as it shows that, thanks to CCRIP

support, farmers were better able to sell their produce despite excessive rainfall which otherwise

damages roads and markets.

Insights from the qualitative data support this finding, with widespread reports that households were

able to increase their income because the markets were accessible and useable all year round, and the

markets were much better attended. In addition, the capacity building support provided to the MMCs

by CCRIP was reported to have greatly improved the sustainable management and maintenance of

the markets, with the increased attendance also providing higher income for the MMCs to perform

its duties.

The improved market access is also reflected by the findings that, although beneficiary households

were not growing significantly more than control households, their production was much more

market oriented, with improved selling practices. It may also have been the case that improved

market access did not stimulate increased production as households did not have to produce more to

account for crop losses caused by delays in accessing markets to sell produce. As a result,

beneficiary households achieved higher incomes from a similar amount of harvest compared to

control households. The fact that this has been achieved without a proportional increase in crop

production overall, combined with the finding that gross margins per hectare has improved only for

the sample within 1km of the market, suggests that benefits from improved input access due to

CCRIP's work may not have applied as much to more remote households compared to the crop sale

benefits.

38

Table 11: Impact of CCRIP on agricultural productivity and sales

2km from market 1km from market

ATE* Obs ATE Obs

Value of ag. production per ha:

Full year 14.2 1,876 9.5 866

Dry seasons only 93.9*** 1,876 16.3 866

Monsoon season only -13.0 1,876 15.8 866

Gross margins per ha. 2.0 1,876 8.7* 866

Proportion of harvest sold 4.8*** 1,876 2.9 866

Sold at market (%

likelihood) 11.0*** 1,269 15.2*** 598

Income from crop sale per ha.:

Full year 104.0*** 1,876 108.3*** 866

Dry seasons 130.2*** 1,876 67.6 866

Monsoon season 69.8** 1,876 65.6 866

Nr. crop varieties (count) -0.02 1,876 -0.1 866

Grew at least one high-value

crop (% likelihood) 7.6*** 1,876 1.3 866

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels,

respectively.

*Coefficients represent percentage change unless specified otherwise.

Table 12 presents the impact of CCRIP on agricultural input use. For the 2km catchment area, we do

not find any significant impacts on the volume of inputs used, including for the amount of land

cultivated, or the amount of seeds, fertiliser, crop protection or labour used during the 12 month

cropping period. As shown in Table 12, we find that gross margins were not significantly improved

amongst beneficiary households for the 2km area, and this is reflected by a lack of a significant impact

on the input productivity (the amount of output produced by one unit of input) of fertiliser, crop

protection, and labour. Only in the case of seed use do we find a positive and signficiant effect on

efficiency, with an increase of 11 per cent in the units of output produced by a unit of input.

As with the results from Table 11, the project's impact on input use intensifies in some cases when we

focus on households located within 1km of the market. Amongst these households, we find that land

use increased significantly, by around 1 hectare, as did the amount of fertiliser and crop protection that

was used (by 53 per cent and 65 per cent respectively). In addition, the input productivity of seeds

increased significantly by 16 per cent, as opposed to 11 per cent for the 2km area. However, as with the

2km area, there was still no significant impact on the productivity of fertiliser, crop protection or labour.

CCRIP was expected to improve households' access to agicultural inputs. In terms of the amount of

inputs that were used, this was not significantly increased for the 12 month period, something which

may have contributed to the lack of a significant impact on crop production. Based on the slightly more

favourable impacts for the 1km catchment area, this lack of impact particularly applied to more remote

households. It was widely reported in the qualitative data that CCRIP markets improved access to

agricultural inputs, but that households often lacked capital to invest in these inputs. Based on this, it

39

may have been the case that physical access to inputs was improved by the project, but additional

barriers such as a lack of capital served to curtail impact in this area.

sThrough additional seasonal analysis, we find that input use was significantly increased for seasonal

crops during the dry season, which again can be linked to the finding for crop production for this season

(where we find a significant positive effect). This additional insight suggests that a lack of impact on

the 12 month period is driven by a lack of input access during the monsoon season. Whilst we find that

crop sale was significantly increased during this season—suggesting the project was able to solve the

barriers to output market access posed by the rains during this season—it is possible that it was not able

to solve the barriers to input market access during this season. Although the positive effect on seed

productivity suggests that households may have been able to access better seeds, but not other inputs

Table 12: Impact of CCRIP on agricultural input use

2km from market 1km from market

ATE* Obs ATE Obs

Land cultivated (ha.) 0.5 1,876 0.9* 866

Value of inputs per ha. 12.9 1,876 19.0 866

Seeds 4.8 1,876 20.4 866

Fertiliser 10.1 1,876 53.2* 866

Crop Protection† 8.2 1,876 64.6** 866

Labour 12.5 1,876 33.6 866

Input productivity of:

Seeds 10.9** 1,610 16.1** 750

Fertiliser 4.4 1,525 -11.2 708

Crop Protection 9.2 1,409 7.0 636

Labour 5.9 1,876 -2.6 866

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

† Crop protection includes following inputs: pesticides, herbicides, fungicides, insecticide

ii. Fish and livestock production

Table 13 presents the impact of CCRIP on fish production and sales (for those who produced fish

during the study period), which bare some similarity to the project's impact on crop production and

sale. We find that the value of production for fish producers in the 2km sample was not increased,

but as with crop sales, we find a large positive impact on income from fish sales of 50 per cent. We

do not find a positive impact on the likelihood of fish being sold at market rather than from home or

at the farm gate, which may be due to the small proportion of fish-producing households who

actually sold fish (18 per cent of fish producers). Likely due to the lack of impact on fish production,

we do not find an impact on the amount of fish harvest used for home consumption. We find similar

results in all cases for the 1km sample.

As with crop sales, the results for fish production highlight the effectivenss of CCRIP in improving

market access for selling goods. As well as potentially being due to a lack of appropriate inputs, the

lack of impact on fish production may have also been due to households not having to produce more

40

to account for losses of their fish harvest due to delays in accesing markets to sell them, an impact

that has also been identified for similar IFAD-funded projects (Brett, 2019). One difference from the

project's impact on crop production, however, is that the impacts on fish production and sales do not

differ by distance to the market, suggesting there are specific barriers to crop production and sales

for more remote households that do not apply to fish.

Table 13: Impact of CCRIP on fish production and sale

2km from market 1km from market

ATE* Obs ATE Obs

Gross value of fish production -17.8% 1,526 -2.8 708

Net value of fish inputs -6.6** 1,526 -10.4** 708

Revenue from fish sales 49.5** 1,526 52.5* 708

Sold fish at market (% likelihood) 10.5 272 -8.0 111

Value of fish consumption per

capita -6.2 1,526 -18.8 708

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

Table 14 presents the impact of CCRIP on livestock production and sales (for those involved in

livestock production as an income source). We do not find a positive impact on the value of production

of livestock for livestock producers, or on income from the sales of livestock and livestock products.

However, we do find a significant 62 per cent increase in the milk productivity of cows. This could

potentially be due to beneficiaries having better access to inputs that could improve cow's milk

productivity, or access to more productive calves and cows themselves. In addition to these results, we

also find that households in the 2km sample were more likely to have consumed meat in the past seven

days.

These results suggest that the impact of improved markets and roads on the sales of crop and fish

produce did not apply to livestock, and thus did not serve to enhance livestock rearing as a livelihood

option. Livestock are usually traded at dedicated markets and livestock trading at the community

markets that were improved by CCRIP is not very common as they are usually taken to larger markets

in more distant locations. In the qualitative data, a common request for future projects was training and

other support for livestock rearing, suggesting that low capacity amongst beneficiaries for livestock

rearing may have also hindered the project's impact in this area.

41

Table 14: Impact of CCRIP on livestock rearing and sale

2km from market 1km from market

ATE* Obs ATE Obs

Gross value of livestock production 10.2 2,171 15.8 1,006

Net value of livestock production 0.9 2,171 1.3 1,006

Revenue from livestock and product sales 12.4 2,171 19.9 1,006

Sold livestock or livestock product (%

likelihood) 1.2 2,171 2.0 1,006

Milk productivity per cow 62.0* 888 -9.4 404

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

iv. Total income and livelihood composition

Table 15 presents the impact of CCRIP on household income and livelihood composition. For the

2km catchment area, we find that CCRIP significantly increased household's total income by 11 per

cent. In addition, beneficiary households were found to be four per cent more likely to be above the

poverty line. We also find that income for wage workers increased by 18 per cent for the 2km

catchment area. Despite these improvements, we do not see an impact of the project on net income,

which may be due to households investing more in their long term production capacities increasing

their expenditures. We also do not find an impact on income from household enterprises or the

number of income sources in the 2km sample, indicating that income diversification did not increase.

Looking at the composition of household income in the 2km catchment area, we see that the

contribution from fish sales was the only income source whose proportion increased significantly,

rising by around 1 percentage point. The only other significant change in income composition is a

significant reduction in remittances (by 2 percentage points), thus suggesting that beneficiary

households maintained their income generating portfolio while increasing total income, hence

needing less remittances.

For households within the 1km catchment area, incomes became more concentrated around crop and

fish sales, reflecting the large impact on these sources. Despite these increases, we see a lower

impact of the project on total household income compared to the 2km sample due to a significant 4

percentage point reduction in the contribution of income from household enterprises, and a

significant 0.8 percentage point reduction in land rental income. In terms of both livelihood diversity

and poverty reduction, however, the results for this sample are more positive than for the 2km

sample. Households within 1km of CCRIP markets are around 5 per cent less likely to be poor, and

there was a slight but significant increase (averaging 0.3) in the number of income sources.

For the 1km catchment area, it is perhaps surprising that total income did not significantly increase

based on the positive impacts on crop and fish sale income for these households. This is likely due to

the fact that crop and fish sales account for just 14 per cent of total household income for these

households, whilst waged labour and household enteprises account for around 38 per cent (see

Section 3b, Table 10). The benefits for these households from the project therefore applied to a more

minor area of their livelihoods, whilst the primary areas were not significantly improved, thus

leading to a curtailed impact on total income.

42

Table 15: Impact of CCRIP on income and livelihood composition

2km from market 1km from market

ATE* Obs ATE Obs

Gross income per capita 10.5** 2,740 9.0 1,312

Net income per capita 0.01 2,740 -1.02 1,312

Income from waged labour

per capita 18.3*** 1,291 11.2 607

Income from household

enterprise per capita -9.1 847 12.7 423

Nr. income sources (count) 0.1 2,740 0.3** 1,312

Above $1.90/ day pov. line

(% likelihood) 4.4** 2,740 5.3* 1,312

Proportion of income from

(percentage points):

Crop sale 1.4 2,740 5.4*** 1,312

Fish sale 0.9** 2,740 1.4** 1,312

Livestock sale -1.4 2,740 -0.9 1,312

Formal waged labour 2.3 2,740 1.2 1,312

Casual waged labour -0.6 2,740 -1.2 1,312

Household enterprises -0.7 2,740 -4.3* 1,312

Remittances -2.0* 2,740 -2.1 1,312

Land rental -0.4 2,740 -0.8** 1,312

Other sources -0.2 2,740 0.6 1,312

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

v. Assets, food security and education

Table 16 presents the impact of CCRIP on household assets, food security and education indicators.

We find that household durable asset ownership was significantly increased by the project in both

catchment areas (the impact is not significant for productive assets only). The impact for ownership

of livestock measured in TLU is not significant for the 2km sample, but slightly negative and

significant for the 1km sample. Regarding food insecurity experience indicators, we find that the

project significantly decreased food insecurity experience both using the full FIES score and

components. Beneficiaries were at least 10 per cent less likely to have worried about having enough

food, and to have been unable to eat healthy food due to a lack of resources during the past year.

Impacts on dietary diversity score and school enrolment are not statistically significant for the 2km

sample.

As with other areas, we find some evidence of more favourable impacts for the 1km sample. For

these households, there was a larger significant increase in the ownership of household assets and a

larger reduction in food insecurity concerns. Although there was also a small and significant

decrease in the dietary diversity score, while the impact on proportion of school-age children

enrolled in school is not significant.

43

The increase in household assets suggests that households may be purchasing assets as a form of

saving and risk mitigation, as opposed to keeping cash savings, which we found was not improved

for the 2km sample. This potentially has positive implications for these households, whose livlihoods

are plagued by a variety of shocks. This finding – combined with the finding of increased income

diversification for the 1km sample – suggests that, by improving the income of beneficiaries, the

project has improved the resilience of households, which has positive implications for the long-term

sustainability of the project impacts.

The contrasting results of the project on food security and dietary diversity (a proxy for nutrition)

may be explained by the high prevalence of poverty within the sample as highlighted in Section 3.

Concerns about having enough food is an issue that applies to the poorest households, who lack

sufficient income to provide for their basic needs, whilst dietary diversity is the concern that follows

once a household has achieved their basic requirements for sustenance. Based on this, the results

suggest that the project was able to help poor households to better meet their basic food needs, but

had less success in helping relatively wealthier households, who were already food secure, to

enhance their dietary diversity.

Table 16: Impact of CCRIP on assets and food security

2km from market 1km from market

ATE Obs ATE Obs

Asset indices:

Household durable

assets 0.1** 2,740 0.1** 1,312

Productive assets -0.02 2,740 0.04 1,312

TLU -0.1 2,740 -0.1* 1,312

FIES score -13.1%*** 2,740 -18.0%*** 1,312

Worried about having

enough food (% likelihood) -9.9*** 2,740 -14.8*** 1,312

Unable to eat healthy food

(% likelihood) -10.7*** 2,740 -13.6*** 1,312

Dietary Diversity Score -0.1 2,740 -0.5*** 1,312

School-age children enrolled

(percentage points) -0.4 2,204 1.8 1,056

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1%

levels, respectively

vi. Financial inclusion

Table 17 presents CCRIP's impact on indicators of financial inclusion. Smallholder farmers across

the world face substantial barriers to obtaining formal bank accounts for payments or savings, as

well as accessing loans (Anderson et al., 2016). Only around 21 per cent of households held a

formal account at the time of the questionnaire, and 20 per cent held another type of account.

Although the impact on the likelihood of having a formal bank account for the 2km sample is not

significant, for the 1km sample we find that the project improved this likelihood by around six per

cent. We also find that the likelihood of having an account other than a formal bank account has

decreased by around seven per cent for the 2km sample.

44

The lack of impact on financial inclusion for the 2km sample is perhaps explained by the lack of

direct focus by the project on improving financial inclusion. As shown in the Theory of Change,

CCRIP's direct impacts were expected through improved market access, and although the results

show that this indeed was improved in terms of buying and selling goods, it does not seem to have

applied to accessing financial services such as bank accounts, savings and loans. It may perhaps be

the case that the institutions that provide these services are based further away than the community

markets that were improved by CCRIP, which we cannot test with our data. Nevertheless, a lack of

access to credit was cited as a key barrier to improving livelihoods in the qualitative data, suggesting

that future projects may benefit from targeting this constraint.

Improved financial inclusion and savings may have also been expected to improve as an indirect

result of higher income stimulated by the project. As we have seen above, CCRIP had a significant

impact on household asset ownership, which is also a common form of saving amongst smallholders

in the country (Anderson et al., 2016). These findings together suggest that households may indeed

be saving more as a result of the income improvements induced by CCRIP, but this is in the form of

assets rather than cash, potentially as a result of persisting issues of access to formal financial

services.

Table 17: Impact of CCRIP on financial inclusion

2km from market 1km from market

ATE* Obs ATE Obs

Have formal bank account

(% likelihood) 2.8 2,740 5.7** 1,312

Have other account

(MFI, NBFI, mobile money)

(% likelihood)

-6.7*** 2,740 -0.8 1,312

Took a loan in past year (%

likelihood) -1.03 2,740 -5.3 1,312

Savings per capita -6.1 2,740 18.6 1,312

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

vii. Women's empowerment

Table 18 presents the impact of CCRIP on indicators of women's autonomous income generation and

their involvement in household decision making. We find that, regardless of the catchment area, the

project did not have a significant effect on the proportion of household income contributed by

women's activities, or on the likelihood of a female household member owning their own enterprise.

We also do not find a significant impact on women's involvement in household decision making

regarding household purchases, children's education or agricultural production or marketing.

Regarding decision making on education, around 83 per cent of the treatment sample reported that

female household members were involved either individually or jointly with male members,

suggesting that there was limited room for improvement in this area. However, there was much-

needed improvement in the other areas, with only 40 per cent of households having women involved

in household purchase decisions, and 58 per cent in terms of agricultural production. In addition, an

45

average of only two per cent of household income is contributed by women's autonomous activities,

and only two per cent of treatment households have a woman owning her own enterprise.

Overall there is little evidence from these results that CCRIP's market and road infrastructure

strengthening helped to improve women's empowerment in terms of autonomous income generation

and household decision making. This may be because these activities of CCRIP did not specifically

target women. The parts of the project that did target women were its use of LCS and the

establishment of women's market areas. As these activities were not expected to reach as many

households as the market and road infrastructure support, it was not feasible to design the household

survey sample to measure the impact of these activities, therefore there may have been an impact of

the project on women's empowerment that is not detected by this assessment. From the qualitative

data, it was reported that the LCS provided a valuable means of gaining income and training for its

members, and served to improve women's standing in the community. However, it was also reported

that some women were forbidden from joining the LCS by their husbands, and that after the work

with CCRIP had finished, female members had difficulty in obtaining additional employment, and

when they did find work their wages were often lower than men's. An additional survey was

conducted that focused on LCS members which may detect further an impact in this area, but the

results from this survey have not yet been produced.

To further investigate the project's impact on women's empowerment, we conducted an additional

analysis that separated the sample into Muslim and non-Muslim households. Interestingly, we found

that the project had a significant positive effect on women's autonomous income generation and their

decision making involvement for agricultural production and sales for the non-Muslim sub-sample.

This suggests that women from Muslim households in these areas may be facing barriers that the

project was unable to overcome. The project has not targeted these potential barriers, as the project's

theory of change assumed that these would not hinder the project's impact (see Section 1b), which is

an assumption that did not hold as these results suggest. Past research has found that, due to

patrilineal and patrilocal kinship systems and traditional gender norms, women face specific barriers

related to their freedom of movement and role in the household, as well as lack control over

economic resources such as assets and credit (Goetz and Sen Gupta, 1996; Donno and Russett, 2004;

Ambler et al., 2017). These findings underline the importance of a thorough understanding of the

general social context surrounding women's participation in the economy and the society in order to

provide improved support to women in similar settings through future projects.

46

Table 18: Impact of CCRIP on women's empowerment

2km from market 1km from market

ATE Obs ATE Obs

Household income from women's

activities (percentage points) 0.6 2,730 0.3 1,308

Women own enterprise (% likelihood) 0.4 2,730 -0.2 1,308

Women involved in decision making (joint or individual) for (% likelihood):

Household purchases 1.5 2,730 1.8 1,308

Children's education -0.2 2,730 -0.4 1,308

Crop or livestock prod. -1.6 1,894 1.3 859

Crop or livestock sales -4.7* 1,894 -2.1 859

b. Impact heterogeneity

Impact by geographic location

Table 19 presents CCRIP's impact on key indicators by geographic location, showing results for

households based in the Barisal, Patuakhali, Bhola and Barguna districts of the Barisal division (the

main area covered by the project), and for households located in the Khulna and Dhaka divisions. The

districts of Khulna and Dhaka could not be analyzed separately due to the smaller sample sizes for these

divisions.

We find that CCRIP's impact varied significantly across these areas. In terms of total income, there was

a significant impact of 60 per cent in Bhola district of Barisal, and a significant impact of 21 per cent in

the Dhaka division. In these areas we also find that households are significantly less likely to be poor

(by 18 and 13 per cent). However, the impact on total income in other areas was not significant.

These differences in impact on total income are shaped by varying impacts on the main income

components. For instance, in Barisal and Patuakhali districts, income from wage labour increased

significantly, as did income for those selling fish in Patuakhali, but a lack of impact in other income

components meant that there was no impact on total income. Income from wage labour significantly

increased also in Khulna, as did income for those selling fish or livestock, but a lack of impact on crop

income or income from household enterprises meant again that total income did not increase. For

Barguna, very large impacts were achieved for those selling crops, fish, and livestock, but a large

significant decrease in income from household enterprises curtailed the project's impact on total

income in this district. In Dhaka, the significant increase in total income, and reduction in poverty, was

driven by a large increase of 209 per cent in livestock income. For Bhola, where the largest impact on

total income was found, the impact was driven by large increases in income from fish sales and wage

labour.

Households in Barguna have the lowest average income at the district level, followed by households in

Khulna and Patuakhali, whilst households in Bhola, Barisal district, and Dhaka have relatively higher

47

incomes. Whilst we find significant increases in total income in Bhola and Dhaka, the promising

findings for Barguna suggest that factors other than wealth may have served to shape the income effect

across these geographic groups. For instance, the improvements in livestock income that drove the

income impact in Dhaka may have been due to an existing facilitating environment for livestock

marketing in this area. This is supported by secondary data showing that livestock ownership in Dhaka

is higher than in Khulna and Barisal (BBS, 2008). Dhaka is also located further inland, so may have

faced less vulnerability to climatic shocks compared to the coastal areas. In Barguna, better access to

markets seem to have pulled households away from household enterprise activities towards crop,

livestock and fish production and sales. The resulting reduction in income from eneterprises, however,

was not fully accounted for by improvements in income from these other sources. Finally for Khulna,

the poorest of the three divisions, there were clearly barriers to improving income from crop sales that

the project was unable to fully overcome, and which hindered the impact on total income as a result.

48

Table 19: Impact of CCRIP by districts of Barisal and Khulna and Dhaka divisions

Barisal Patuakhali Bhola Barguna Khulna districts Dhaka districts

ATE* Obs ATE Obs ATE Obs ATE Obs ATE Obs ATE Obs

Gross income per capita -9.1 427 -11.9 432 60.1*** 407 15.3 426 -1.7 454 21.0** 594

Above $1.90/ day pov. line (%

likelihood) -6.0 427 2.9 432 18.3*** 407 -3.8 426 -0.8 454 12.6** 594

Income from crop sale per ha. -58.8 334 34.2 286 15.4 338 363.7*** 320 -13.8 212 20.4 390

Income from fish sale -55.4 173 170.1** 214 158.7*** 298 124.9*** 373 164.1** 345 -88.4 123

Income from livestock sale -76.5 138 -110.7* 358 12.3 344 107.7** 395 120.8* 341 208.7*** 398

Income from waged labour per

capita 75.1*** 138 38.4*** 255 52.1*** 218 -5.5 194 26.6*** 304 3.3 182

Income from household

enterprise per capita -19.9 114 -25.5 130 61.6 142 -108.6*** 93 24.9 139 -24.6 229

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

ii. Impact by catchment area

Table 20 presents CCRIP's impact on key indicators for households located within 2km of the

CCRIP market but not within 1km of the connecting road, and for households who are located both

within 2km of the market and within 1km of the connecting road. Somewhat surprisingly, we find a

larger impact on total income for those not located close to the road (13 per cent), and a non-

significant impact for those located closer to the road. We also find that households that are only

within a market catchment area (i.e. further than 1km of a road) were less likely to be poor. These

impacts on those located further from the road were driven by a larger impact on income from

selling crops, which averaged 150 per cent compared to 81 percent for those located closer to the

road, and a 69 per cent increase in income from livestock sales. Housheolds located closer to the

road observed a larger increase in income from fish sales and wage labour, but no significant

increase in total income.

Households who are located close to both the CCRIP market and the connecting road were

theorectically expected to benefit the most from the project as they were receiving two types of

support. However, our results show that those located further from the connecting road benefitted

more in terms of total income. Given that households located further from the road are poorer (see

Section 3b), it may be the case that within the market catchment area, the impact of the project was

more pro-poor. Unlike previous studies of infrastructure investments that found a lack of impact on

the poorest households, in the case of CCRIP poorer households starting from a lower base of

income, and facing more barriers, seemingly experienced larger increases thanks to the project's

support (Khandker and Koolwall, 2010). In addition, these results also suggest that the connecting

roads were perhaps more important for certain livelihood activities, such as the fish sales and wage

labour, than for the sale of crops or livestock products.

Table 20: Impact of CCRIP by catchment area

Market only Market and road

ATE* Obs ATE Obs

Gross income per capita 12.7* 1,491 7.5 1,249

Above $1.90/ day pov. line

(% likelihood) 5.3* 1,491 3.8 1,249

Income from crop sales 150.4*** 985 81.2** 891

Income from fish sales 59.6* 865 70.7** 661

Income from livestock sales 69.3** 1,170 -32.9 1,001

Income from wage labour

per capita 17.0 768 18.4** 523

Income from household

enterprise per capita -26.6 457 5.7 390

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

iii. Impact by livelihood activities

The whole sample used for this analysis includes households that are primarily farm households, and

those that rely primarily on other income generating activities. In order to assess whether CCRIP

impact varied by these two household groups, Table 21 presents impacts on the income of

households who were involved in farming (those who sold any crops, produced any fish, or kept any

livestock during the 12 month study period), and for those households not involved in farming. We

find that the project's impact is driven by improvements amongst farming households, with their total

income increasing significantly by 16 per cent, compared to a negligible impact for non-farming

households. Along with increased farm income, the impact on total income for farming households

was also driven by higher income for wage labourers. As a result of these impacts, farming

households were 6 per cent less likely to be poor, while no impact was found for non-farming

households.

The average income for farming households is around 30 per cent lower than the average income of

non-farming households, and farming households also are more likely to be below the poverty line.

The fact that the project only impacted farming households provides further evidence that the project

was more beneficial for poorer households within the market catchment area. It should also be

considered, however, that project activities may not have been suited to the livelihood needs of non-

farming households in the project areas, with markets seemingly being beneficial for sales of crops,

fish, and in rare cases livestock, but not for providing opportunities for household enterprises and, in

some cases, wage labour. This insight is also supported by the income impacts for households

located within the 1km catchment area (see Section 4a-iv).

Table 21: Impact of CCRIP by livelihood activities

Farm households Non-farm households

ATE* Obs ATE Obs

Gross income per capita 15.9*** 2,163 -6.7 577

Above $1.90/ day pov. line

(% likelihood) 5.5** 2,163 -0.9 577

Wage income per capita 14.2** 991 14.03 300

Household enterprise

income per capita 3.0 648 -20.9 199

5. Conclusion

The Coastal Climate Resilient Infrastructure Project aimed to improve the livelihoods of poor and

remote households in Southwest Bangladesh through improved markets and roads. In particular, the

project aimed to improve the access and useability of community markets during seasonal flooding.

Using data from an in-depth household questionnaire combined with qualitative interviews, we have

rigorously assessed the project's impact on a range of impact indicators relating to income; crop, fish

and livestock production and sales; assets, food security and education; financial inclusion; and

women's empowerment. We have assessed impact on the whole sample, as well as for a range of

sub-groups, including by geographic location, location within the market catchment area, and by

livelihood activity, integrating findings from the qualitative data to help to explain the mechanisms

that shaped the project's impact.

Overall, we have found that this type of focused infrastructure project can improve the climate

resilience and accessibility of local markets, leading to significant improvements in income, assets

and food security. This is achieved by improving households' market access in both the dry and

monsoon seasons, and thus the marketing practices of beneficiaries, particularly in terms of crop and

fish sales. As well as infrastructure support, we find that these impacts were facilitated through

capacity building support to the local institutions managing the markets, which contributed to the

sustainable management of the project markets. Through sub-group analyses we find that this type of

project can be particularly beneficial for poorer, more remote households, although some impacts,

such as on livestock income and wage labour vary depending on facilitating conditions.

A number of lessons can be drawn from these findings to improve the impact of future climate

resilient infrastructure projects. First, future projects should pay special attention to ensuring

households have access to high-quality agricultural inputs, as improved access to output markets

does not necessarily improve access to inputs especially for capital constrained households.. In

addition to improving input access, agricultural productivity and income could also be improved by

providing complementary training and technology support.

Second, future projects should consider different components of beneficiaries' livelihoods and

provide activities to stimulate the main sources of income, which may vary for different areas. In

some cases for CCRIP, there was a lack of impact on the main income sources for some households

(mainly wage labour and household enterprises), leading to a lack of impact on total income. Impact

on income could thus be enhanced by offering complementary support for the livelihood activities

that are the most important in each local context. In the case of wage labour, this could involve a

redesign of the LCS activities to ensure that the valuable employment and training provided does

not remain short-term in nature and includes support to establish linkages with local labor market

for sustained impacts. In terms of household enterprises, these activities could be improved by

facilitating easier entry into local markets for small shops and traders, as well as providing credit and

training for setting up and managing these businesses.

Finally, future projects should provide more extensive support to improve the income generating

opportunities and overall empowerment of women, especially in countries such as Bangladesh where

women face ingrained barriers to their mobililty and autonomy. Although the LCS impacts for

CCRIP were promising, these activities combined with market and road support that did not

explicity target these barriers were largely insufficient for improving women's empowerment,

particularly in Muslim households. Based on the success of initiatives such as BRAC's

Empowerment and Livelihood for Adolescents program in Bangladesh (as well as Afghanistan,

Haiti, Sierra Leone, South Sudan, Tanzania and Uganda - see Bandiera et al. 2018) future projects

could provide multi-faceted support to improve the hard and soft skills of women, provided within a

safe space environment, and involve the wider society to ensure sustainability of impacts.

References

Ahmed, A.U., Ahmad, K., Chou, V., Hernandez, R., Menon, P., Naeem, F., Naher, F., Quabili, W.,

Sraboni, E. and Yu, B. 2013. The status of food security in the Feed the Future Zone and other

regions of Bangladesh. Paper Prepared for USAID Grant No. EEM-G-00-04-00013-00. Dhaka,

Bangladesh: IFPRI.

Ahmed, Z. 2010. Role of local government in indigienous market management in the rural areas of

Bangladesh: Do these markets play development roles? Journal of Sustainable Development, 3(1):

120-135

Alkire, S. and M.E. Santos. 2010. Acute multidimensional poverty: A new index for developing

countries. Oxford Poverty and Human Development Initiative (OPHI) Working Paper No. 38.

Oxford, UK: OPHI. https://www.econstor.eu/bitstream/10419/48297/1/3_alkire.pdf

Anderson, J., Moler, A. and Kretchun, N. 2016. National survey and segmentation of smallholder

households in Bangladesh: Understanding their demand for financial, agricultural, and digital

solutions. CGAP Working Paper. Washington, D.C., USA: CGAP.

Arslan, A., Cavatassi, R., Alfani, F., McCarthy, N., Lipper, L., and Kokwe, M. 2018a.

Diversification Under Climate Variability as Part of a CSA Strategy in Rural Zambia, The Journal of

Development Studies, 54 (3): 457-480.

Arslan, A., Higgins, D. & Winters, P. 2018b. Impact Assessment Report: Integrated Rice Production

Enhancement Project (IRPEP), The Philippines. Rome, Italy: IFAD. Available at:

https://drive.google.com/file/d/1VRVKDvdkJedsrE9-ncH1HiLUgGS7nmLI/view?usp=sharing

Austin, P. C. 2009. Balance diagnostics for comparing the distribution of baseline covariates

between treatment groups in propensity-score matched samples. Statistics in Medicine, 28(25): 3083-

3107

Austin, P.C. 2011. An introduction to propensity score methods for reducing the effects of

confounding in observational studies. Multivariate Behavioural Research, 46(3): 399-424

Austin, P.C. and E.A. Stuart. 2015. Moving towards best practice when using inverse probability of

treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in

observational studies. Statistics in Medicine, 34: 3661-3679

Ballard, T., Kepple, A.W. and Cafiero, C. 2013. The Food Insecurity Experience Scale:

Development of a global standard for monitoring hunger worldwide. FAO Technical Paper. Rome,

Italy: FAO.

Bandiera, O., N. Buehren, R. Burgess, M. Goldstein, S. Gulesci, I. Rasul and M. Sulaiman. 2018.

Women’s Empowerment in Action: Evidence from a Randomized Control Trial in Africa. London:

Economic Organization and Public Policy

Barrett, C. 2008. Smallholder market participation: Concepts and evidence from eastern and

southern Africa. Food Policy, 33: 299-317

Baser, O. 2006. Too much ado about Propensity Score Models? Comparing methods of Propensity

Score Matching. Value in Health 9(6): 377-385.

Bokelmann, W. And M.E. Adamseged. 2016. Contributing to a better understanding of the value

chain framework in developing countries. Invited poster presented at the 5th International

Conference of the African Association of Agricultural Economists, September 23-26, 2016, Addis

Ababa, Ethiopia.

Caliendo, M. and S. Kopeinig. 2008. Some practical guideance for the implementation of propensity

score matching. Journal of economic surveys, 22(1):31-72

Carletto, G. Covarrubias, K. Davis, B., Krausova, M. and Winters, P. 2007. Rural Income

Generating Activities Study: Methodological note on the construction of income aggregates.Rome,

Italy: FAO.

Carletto, G., Ruel, M., Winters, P. & Zezza, A. 2015. Farm-level pathways to improved nutritional

status: Introduction to the special issue. Journal of Development Studies, 51: 945-957.

Cropper, M. and Griffiths, C. 1994. The interaction of population growth and environmental quality,

The American Economic Review, 84(2): 250-254.

Dehejia, R.H. and Wahba, S. 1999. Causal effects in nonexperimental studies: Reevaluating the

evaluation of training programs. Journal of the American Statistical Association, 94(448): 1053-

1062.

Donno, D. and Russett, B. 2004. Islam, authoritarianism, and female empowerment: What are the

linkages? World Politics, 56(4): 582-607.

Ellis, F. 1999. Rural livelihood diversity in developing countries: Evidence and policy implications.

ODI Natural Resource Perspectives No. 40. https://www.odi.org/sites/odi.org.uk/files/odi-

assets/publications-opinions-files/2881.pdf

Fan, M., J. Shen, L. Yuan, R Jiang, X. Chen, W.J. Davies, F. Zhang. 2012. Improving crop

productivity and resource use efficiency to ensure food security and environmental quality in China.

Journal of Experimental Botany, 63(1): 13-24

FAO (Food and Agriculture Organisation of the United Nations). 2003. Planning and designing

rural markets. Rome, Italy: FAO.

Ferreira, F.H.G., Chen, S., Dabalen, A.L., Dikhanov, Y.M., Hamadeh, N., Jolliffe, D.M., Narayan,

A., Prydz, E.B., Revenga, A.L., Sangraula, P., Serajuddin, U., Yoshida, N. 2015. A global count of

the extreme poor in 2012: Data issues, methodology and initial results. Policy Research Paper No

WPS 7432. Washington, DC., USA: World Bank

Filmer, D. and Pritchett, L.H. 2001. Estimating wealth effects without expenditure data – or tears:

An application to educational enrolments in states of India. Demography, 38(1): 115-132

Gallo, A. 2016. A refresher on statistical significance. Harvard Business Review.

https://hbr.org/2016/02/a-refresher-on-statistical-significance

Garbero, A. 2016. Measuring IFAD's impact: Background paper to the IFAD9 Impact Assessment

Initiative. IFAD Research Series Issue 7. Rome, Italy: IFAD

Goetz, A.M. and Sen Gupta, R. 1996. Who takes the credit? Gender, power, and control over loan

use in rural credit programs in Bangladesh. World Development, 24(1): 45-63.

Gulati, A., N. Minot, C. Delgado, and S. Bora. 2005. Growth in high-value agriculture in Asia and

the emergence of vertical links with farmers. Paper presented at the workshop "Linking small-scale

producers to markets: Old and new challenges" The Work Bank, 15 December 2005.

Hetherington, J.B., Wiethoelter, A.K., Negin, J. & Mor, S.M. 2017. Livestock ownership, animal

source foods and child nutritional outcomes in seven rural village clusters in Sub-Saharan Africa.

Agriculture and Food Security, 6(9): 1-11.

Huq, N., J. Huge, E. Boon, and A.K. Gain. 2015. Climate change impacts in agricultural

communities in rural areas of coastal Bangladesh: A tale of many stories. Sustainability, 7:8437-

8460

Islam, M.S., Sultana, S., Saifunnahar, M. and Miah, M.A. 2014. Adaptation of char livelihood in

flood and river erosion areas through indigenous practice: A study on Bhuapur Riverine Area in

Tangail. Journal of Environmental Science and Natural Resources, 7(1): 13-19.

Jahnke, H.E. 1982. Livestock production systems and livestock development in tropical Africa. Kiel,

Germany: Kieler Wissenschaftersverlag Vauk

Jones, L., Jaspars, S., Pavanello, S., Ludi, E., Slater, R., Arnall, A., Grist, N. and Mtisi, S. 2010.

Responding to a changing climate: Exploring how disaster risk reduction, social protection and

livelihoods approaches promote features of adaptive capacity. ODI Working Paper No. 319.

London, UK: Overseas Development Institute.

Khandker, R., Z. Bakht, and G.B. Koolwal. 2009. The poverty impact of rural roads: Evidence from

Bangladesh. Economic Development and Cultural Change, Vol 57(4): 685-722

Khandker, S. R., Koolwal, G. B. & Samad, H. A. 2010. Handbook on impact evaluation:

Quantitative methods and practice. Washington, DC, USA: World Bank.

Kirtman, B., S.B. Power, J.A. Adedoyin, G.J. Boer, R. Bojariu, I. Camilloni, F.J. Doblas-Reyes,

A.M. Fiore, M. Kimoto, G.A. Meehl, M. Prather, A. Sarr, C. Schär, R. Sutton, G.J. van Oldenborgh,

G. Vecchi and H.J. Wang. 2013: Near-term Climate Change: Projections and Predictability. In:

Stocker, T.F., D. Qin, G.-K. Plattner, M. Tignor, S.K. Allen, J. Boschung, A. Nauels, Y. Xia, V. Bex

and P.M. Midgley. (eds.) Climate Change 2013: The Physical Science Basis. Contribution of

Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change.

Cambridge, UK: Cambridge University Press.

Meybeck, A., J. Lankoski, S, Redfern, N, Azzu, and V. Gitz. 2012. Building resilience for

adaptation to climate change in the agriculture sector. Proceedings of a join FAO/OECD workshop

23-24 April 2012. Rome, Italy: FAP

Nishat, A., N. Mukherjee, E. Roberts, and A. Hasemann. 2013. A range of approaches to address

loss and damage from climate change impacts in Bangladesh. Loss and Damage Publication. Bonn,

Germany: Germanwatch

Rahman, S. and M.A. Rahman. 2015. Climate extremes and challenges to infrastructure development

in coastal cities in Bangladesh. Weather and Climate Extremes, 7:96-108

Rehima, M., K. Belay, A. Dawit, S. Rashid. 2013. Factors influencing farmers' crop diversification:

Evidence from SNNPR, Ethiopia. International Journal of Agricultural Studies, 3(8): 558-565

Relief Web. 2008. Cyclone Sidr in Bangladesh: Damage, loss and needs assessment for disaster

recovery and reconstruction. A report prepared by the Government of Bangladesh assisted by the

international development community with financial support from the European Commission. New

York City, USA: Relief Web

Relief Web. 2009. Cyclone Aila losses in Bangladesh estimated at USD269 million.

https://reliefweb.int/report/bangladesh/cyclone-aila-losses-bangladesh-estimated-269-mln-usd

Roy, S.K., F Bilkes, K. Islam, G. Ara, P. Tanner, I. Wosk, A.S. Rahman, B. Chakraborty, S.P. Jolly,

W. Khatun. 2008. Impact of pilot project of Rural Maintenance Programme (RMP) on destitute

women: CARE, Bangladesh. Food and Nutrition Bulletin, 29(1): 67-75

Saha, C.K. 2014. Dynamics of disaster-induced risk in southwestern coastal Bangladesh: An analysis

on Tropical Cyclone Aila 2009. Natural Hazards. 75(1): 727-754

Sarker, M.H., Huque, I., Alam, M. and Koudstaal, R. 2003. Rivers, chars and char dwellers of

Bangladesh. International Journal of River Basin Management, 1(1): 61-80.

Sheoran, J. 2016. Aspects of women empowerment: A brief overview. International Research

Journal of Management, Science and Technology, 7(4):135-140

Svensson, J. and D.Y. Drott. 2010. Tuning in the market signal: The impact of market price

information on agricultural outcomes. Institute for International Economic Studies Working Paper.

Stockholm, Sweden: Stockholm University

Vallejo, L. and M. Mullan. 2017. Climate-resilient infrastructure: Getting the policies right. OECD

Environmental Working Papers No. 121. Paris, France: OECD Publishing

White, H. Theory-based impact evaluation: Principles and practice. Journal of Development

Effectiveness, 1(3): 271-284.

Wooldridge, J.M. 2010. Econometric analysis of cross section and panel data. 2nd

ed. Cambridge,

MA: MIT Press

World Bank. 2001. Engendering Development: A World Bank Policy Research Report. Washington,

DC., USA: The World Bank

World Bank, 2017. Atlas of sustainable development goals 2017: From World Development

Indicators. Washington, DC, USA: World Bank

World Bank, 2017. GDP per capita (current US$), Bangladesh. [Online] Available at:

https://data.worldbank.org/indicator/NY.GDP.PCAP.CD

Appendix I: Mean values for impact indicators

I.A: Treatment and control means for impact indicators of agricultural productivity and sales

Treatment Control Diff.

Mean Obs Mean Obs

Value of ag. production per ha (BDT):

Full year 149,466 806 43,160 1,070 106,306*

Dry seasons only 59,452 806 39,944 1,070 19,508***

Monsoon season only 38,666 806 41,020 1,070 -2,354

Gross margins per ha (BDT). 111,945 806 10,632 1,070 101,313*

Proportion of harvest sold (%) 33.2 806 27.1 1,070 6.1***

Sold at market (%) 50.3 586 41.6 683 8.7***

Income from crop sale per ha. (BDT):

Full year 40,879 806 22,455 1,070 18,424***

Dry seasons 31,577 806 17,267 1,070 14,310***

Monsoon season 48,351 806 26,926 1,070 21,425***

Nr. crop varieties (count) 2.6 806 2.6 1,070 0

Grew at least one high-value crop

(%) 43.9 806 38.3 1,070 5.6**

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

I.B: Treatment and control means for impact indicators of agricultural input use

Treatment Control Diff.

Mean Obs Mean Obs

Land cultivated (ha.) 3.3 806 2.8 1,070 0.5*

Value of inputs per ha. (BDT) 37,522 806 32,528 1,070 4,994**

Seeds 6,457 806 3,606 1,070 2,851**

Fertiliser 4,653 806 4,054 1,070 599

Crop Protection† 3,519 806 3,717 1,070 -198

Labour 27,543 806 25,585 1,,070 1,958

Input productivity of (BDT):

Seeds 66.9 710 40.4 900 26.5

Fertiliser 774.3 671 29.5 854 744.8

Crop Protection 25,970 629 119.6 780 25,850

Labour 329.6 806 201.9 1,070 127.7*

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

I.C: Treatment and control means for impact indicators of fish production and sale

Treatment Control Diff.

Mean Obs Mean Obs

Gross value of fish production

(BDT) 33,181 626 26,746 900 6,435

Net value of fish inputs (BDT) 241,260 626 235,708 900 5,552

Revenue from fish sale (BDT) 18,573 626 5,429 900 13,144

Sold fish at market (%) 55.0 129 41.3 143 13.7**

Value of fish consumption per

capita (BDT) 2,390 626 2,201 900 189

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

I.D: Treatment and control means for impact indicators of livestock rearing and sale

Treatment Control Diff.

Mean Obs Mean Obs

Gross value of livestock

production (BDT) 40,477 944 31,490 1,227 8,987

Net value of livestock production

(BDT) 410,993 944 404,329 1,227 6,664

Revenue from livestock and

product sale (BDT) 30,734 944 22,811 1,227 7,923

Sold livestock or livestock

product (%) 52.4 944 50.3 1,227 2.1

Milk productivity per cow (BDT) 8,342 356 6,218 532 2,124***

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

I.E: Treatment and control means for impact indicators of income and livelihood composition

Treatment Control Diff.

Mean Obs Mean Obs

Gross income per capita (BDT) 79,588 1,188 69,177 1,552 10,411

Net income per capita (BDT) 55,658 1,188 45,277 1,552 10,381*

Income from waged labour per

capita (BDT) 51,328 554 35,901 737 15,427

Income from household enterprise

per capita (BDT) 98,508 369 98,628 478 -120

Nr. income sources (count) 7.9 1,188 7.6 1,552 0.3**

Above $1.90/ day pov. line (%) 75.3 1,188 73.3 1,552 2.0

Proportion of income from (%):

Crop sale 16.1 1,188 13.3 1,552 2.8***

Fish sale 2.1 1,188 1.3 1,552 0.8**

Livestock sale 6.2 1,188 6.6 1,552 -0.4

Formal waged labour 22.2 1,188 20.5 1,552 1.7

Casual waged labour 10.0 1,188 11.8 1,552 -1.8*

Household enterprises 23.5 1,188 23.4 1,552 0.1

Remittances 12.7 1,188 16.2 1,552 -3.5***

Land rental 1.1 1,188 1.2 1,552 -0.1

Other sources 4.1 1,188 4.1 1,552 0

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

I.F: Treatment and control means for impact indicators of assets, food security and education

Treatment Control Diff.

Mean Obs Mean Obs

Asset indices:

Household durable assets 1.2 1,188 1.2 1,552 0

Productive assets 0.6 1,188 0.6 1,552 0

TLU 0.8 1,188 0.8 1,552 0

FIES score 2.2 1,188 2.6 1,552 -0.4***

Worried about having enough

food (%) 51.7 1,188 62.8 1,552 -11.1***

Dietary Diversity Score 10.5 1,188 10.6 1,552 -0.1

School-age children enrolled (%) 70.0 952 71.1 1,252 -1.1

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

I.G: Treatment and control means for impact indicators of financial inclusion

Treatment Control Diff.

Mean Obs Mean Obs

Have formal bank account (%) 20.6 1,188 17.8 1,552 2.8*

Have other account

(MFI, NBFI, mobile money) (%) 19.6 1,188 24.4 1,552 -4.8***

Took a loan in past year (%) 51.0 1,188 52.5 1,552 -1.5

Savings per capita (BDT) 2,060 1,188 2,018 1,552 42

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

I.H: Treatment and control means for impact indicators of women's empowernment

Treatment Control Diff.

Mean Obs Mean Obs

Household income from women's

activities (% of total income) 1.6 1,182 1.3 1,548 0.3

Women own enterprise (%) 2.1 1,182 2.0 1,548 0.1

Women involved in decision making (joint or individual) for (%):

Household purchases 39.8 1,182 37.7 1,548 2.1

Children's education 82.9 1,182 85.7 1,548 -2.8**

Crop or livestock prod. 57.8 778 61.1 1,043 -3.3

Crop or livestock sales 52.3 778 57.2 1,043 -4.9**

Note: *,** and *** indicate that the difference between the treatment and control means is statistically significant at

the 10, 5 and 1% levels, respectively.

Appendix II: Distribution of Propensity Scores before and after trimming

Before trimming

After trimming

Appendix III: Results from the secondary IPWRA model

III.A: Impact of CCRIP on agricultural productivity and sales

2km from market 1km from market

ATE* Obs ATE Obs

Value of ag. production per ha:

Full year 15.8 1,876 14.8 866

Dry seasons only 148.7** 1,876 122.4** 866

Monsoon season only -24.4 1,876 -4.4 866

Gross margins per ha. 6.0 1,876 5.6 866

Proportion of harvest sold 6.9** 1,876 3.8 866

Sold at market (% likelihood) 6.7 1,269 15.3** 598

Income from crop sales per ha.:

Full year 114.3** 1,876 134.1*** 866

Dry seasons 186.0** 1,876 143.8** 866

Monsoon season 99.2 1,876 79.8 866

Nr. crop varieties (count) 3.1 1,876 -2.3 866

Grew at least one high-value

crop (% likelihood) 4.9

1,876 3.0

866

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

III.B: Impact of CCRIP on agricultural input use

2km from market 1km from market

ATE* Obs ATE Obs

Land cultivated (ha.) -6.8 1,876 76.2 866

Value of inputs per ha. 13.5 1,876 24.4 866

Seeds 9.2 1,876 21.5 866

Fertiliser 5.9 1,876 68.6* 866

Crop Protection† 32.9 1,876 96.0** 866

Labour 15.9 1,876 47.2 866

Input productivity of:

Seeds 7.0 1,610 26.6** 750

Fertiliser 8.3 1,525 -1.4 708

Crop Protection 3.6 1,409 0.8 636

Labour 6.6 1,876 -4.0 866

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

† Crop protection includes following inputs: pesticides, herbicides, fungicides, insecticide

III.C: Impact of CCRIP on fish production and sale

2km from market 1km from market

ATE* Obs ATE Obs

Gross value of fish production -18.1 1,456 2.0 664

Net value of fish inputs -1.3 1,456 -1.9 664

Revenue from fish sales 51.0 1,456 47.0 664

Sold fish at market (% likelihood) 8.6 257 -1.24 101

Value of fish consumption per

capita -14.3 1,456 -17.2 664

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

III.D: Impact of CCRIP on livestock rearing and sale

2km from market 1km from market

ATE* Obs ATE Obs

Gross value of livestock production 9.1 2,171 14.6 1,006

Net value of livestock production -0.7 2,171 1.2 1,006

Revenue from livestock and product sales 22.3 2,171 2.7 1,006

Sold livestock or livestock product (%

likelihood) 2.5 2,171 0.1 1,006

Milk productivity per cow -31.1 2,267 16.0 404

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

III.E: Impact of CCRIP on income and livelihood composition

2km from market 1km from market

ATE* Obs ATE Obs

Gross income per capita 6.0 2,740 7.2 1,312

Net income per capita -0.7 2,740 -2.4 1,312

Income from wage labour

per capita 14.8** 1,291 10.5 607

Income from household

enterprise per capita -4.9 847 8.3 423

Nr. income sources (count) 0.2 2,740 0.4 1,312

Above $1.90/ day pov. line

(% likelihood) 2.6 2,740 5.5 1,312

Proportion of income from (percentage points):

Crop sales 0.6 2,740 5.2** 1,312

Fish sales 0.6 2,740 0.7 1,312

Livestock sales -0.9 2,740 -1.7 1,312

Formal wage labour 2.4 2,740 2.0 1,312

Casual wage labour -1.1 2,740 -2.5 1,312

Household enterprises -0.6 2,740 -3.9 1,312

Remittances -1.4 2,740 -0.6 1,312

Land rental -0.1 2,740 -0.4 1,312

Other sources -0.2 2,740 0.9 1,312

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

III.F: Impact of CCRIP on indicators of assets, food security and education

2km from market 1km from market

ATE Obs ATE Obs

Asset indices:

Household durable

assets 0.1 2,740 15.1** 1,312

Productive assets -0.01 2,740 6.3** 1,312

TLU -0.1 2,740 -12.6 1,312

FIES score -13.8%** 2,740 -23.3%*** 1,312

Worried about having

enough food (% likelihood) -11.2*** 2,740 -17.5*** 1,312

Dietary Diversity Score -0.6 2,740 -39.5 1,312

School-age children enrolled

(percentage points) -1.7 2,204 3.4 1,056

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

III.G: Impact of CCRIP on financial inclusion

2km from market 1km from market

ATE* Obs ATE Obs

Have formal bank account

(% likelihood) 2.2 2,740 2.4 1,312

Have other account

(MFI, NBFI, mobile money)

(% likelihood) -4.2* 2,740 0.2 1,312

Took a loan in past year (%

likelihood) 0.2 2,740 -3.9 1,312

Savings per capita -4.2 2,740 5.1 1,312

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.

*Coefficients represent percentage change unless specified otherwise.

III.H: Impact of CCRIP on women's empowernment

2km from market 1km from market

ATE Obs ATE Obs

Household income from women's

activities (percentage points) 0.3 2,730 0.5 1,308

Women own enterprise (% likelihood) 0.2 2,730 1.1 1,308

Women involved in decision making (joint or individual) for (% likelihood):

Household purchases 3.1 1,821 6.9 1,308

Children's education -1.0 1,821 3.1 1,308

Crop or livestock prod. -0.3 1,821 3.5 838

Crop or livestock sales -2.3 1,821 -0.5 838

Note: *,** and *** indicate that the estimated ATE is statistically significant at the 10, 5 and 1% levels, respectively.